Scholarlink

<p>Today's agricultural practice is the contemporary application of technologies and practices of food production, which produce agricultural products to keep up with global demand for their products. Although practices often do not fully account for the complexities of space and time, especially when assessing the altered relationships of large-scale data representing the environmental atmosphere. This reduction in complexity ultimately affects the statistical models used to predict behavior in environmental communities that have multiple influential factors. To tackle these challenges, this research presents an innovative and cohesive framework known as the Deep Twin Bottleneck Residual Convolutional Neural Network (DTBRCNN), designed for reliable soil classification across various agro-ecological zones, effectively capturing spatial patterns for categorizing different soil types. A Spatial–Temporal Pufferfish Long Short-Term Memory Graph Neural Network (SPLSTMGNN) that models intricate environmental dynamics, creatively combining the temporal modeling strengths with the spatial reasoning capabilities. Furthermore, this integrates the Pufferfish Optimization Algorithm (POA), enhances the parameter optimization, and therefore gives a quick convergence with a minimized risk of ending up in a local minimum. This approach minimizes the error function for a suitable mix of crops for a specific region and season, taking into account various factors like yield, resource utilization and climatic conditions. Finally, the outcomes of both segments are fused using a cross-attention technique for multi-crop recommendation. An explainable artificial intelligence approach, known as Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) approaches, is used to improve model transparency by obtaining explanations of predictions. The proposed approach achieves 99% accuracy and 0.165&#xa0;s latency for crop recommendations. Thus, the proposed integrated framework enhances soil classification and weather prediction in modern agriculture.</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

Hybrid deep learning with a fusion model of weather and soil-based multi-crop recommendations for improving crop sustainability

  • 期刊论文

  • 发表日期：

    2025年11月7日

查看全文
  • Mohan M,
  • S. Anandamurugan

摘要

Today's agricultural practice is the contemporary application of technologies and practices of food production, which produce agricultural products to keep up with global demand for their products. Although practices often do not fully account for the complexities of space and time, especially when assessing the altered relationships of large-scale data representing the environmental atmosphere. This reduction in complexity ultimately affects the statistical models used to predict behavior in environmental communities that have multiple influential factors. To tackle these challenges, this research presents an innovative and cohesive framework known as the Deep Twin Bottleneck Residual Convolutional Neural Network (DTBRCNN), designed for reliable soil classification across various agro-ecological zones, effectively capturing spatial patterns for categorizing different soil types. A Spatial–Temporal Pufferfish Long Short-Term Memory Graph Neural Network (SPLSTMGNN) that models intricate environmental dynamics, creatively combining the temporal modeling strengths with the spatial reasoning capabilities. Furthermore, this integrates the Pufferfish Optimization Algorithm (POA), enhances the parameter optimization, and therefore gives a quick convergence with a minimized risk of ending up in a local minimum. This approach minimizes the error function for a suitable mix of crops for a specific region and season, taking into account various factors like yield, resource utilization and climatic conditions. Finally, the outcomes of both segments are fused using a cross-attention technique for multi-crop recommendation. An explainable artificial intelligence approach, known as Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) approaches, is used to improve model transparency by obtaining explanations of predictions. The proposed approach achieves 99% accuracy and 0.165 s latency for crop recommendations. Thus, the proposed integrated framework enhances soil classification and weather prediction in modern agriculture.