Smart Farming: Enhancing Urban Agriculture Through Predictive Analytics and Resource Optimization

作     者：Aldhahri, Eman Ali Almazroi, Abdulwahab Ali Alkinani, Monagi Hassan Ayub, Nasir Alghamdi, Elham Abdullah Janbi, Nourah Fahad 

作者机构：Univ Jeddah Dept Comp Sci & Artificial Intelligence Coll Comp Sci & Engn Jeddah 21959 Saudi Arabia Univ Jeddah Coll Comp & Informat Technol Khulais Dept Informat Technol Jeddah 21959 Saudi Arabia Air Univ Islamabad Dept Creat Technol Islamabad 44000 Pakistan 

出 版 物：《IEEE ACCESS》 (IEEE Access)

年 卷 期：2025年第13卷

页      面：72375-72388页

核心收录：

基　　金：University of Jeddah  Jeddah  Saudi Arabia [UJ-23-DR-210] 

主　　题：Monitoring Crops Temperature measurement Adaptation models Temperature sensors Plants (biology) Accuracy Predictive models Urban agriculture Deep learning Crop health monitoring environmental stress assessment ResXceNet-HBA data imputation feature selection precision agriculture 

摘      要：Optimal agricultural methods need precise crop health and ecological strain monitoring. This study proposes a novel data science strategy to improve crop health prediction and stress assessment. ResXceNet-HBA is a cutting-edge classification model that uses ResNet blocks, Xception modules with Adaptive Depthwise Separable Convolutions, and HBA-optimized parameters. This model uses HBA s Dynamic Exploration-Exploitation Balance-fine-tuned Dynamic Feature Recalibration and adaptive convolutions. Imputation Weight Crop Labels (WICL) to accurately fill in missing data, Localised Feature Scaling (LFS) and Adaptive Feature Discretization (AFD) to standardize and categorize features, and the Environmental Stress Factor (ESF) to evaluate crop stress address data problems ASRFS and Crop-Specific Environmental Impact Weighting increase model performance. Our system also employs Adaptive Synthetic Resampling with Environmental Context. Using novel measures including the Crop Type Generalisation Score (CTGS) and Environmental Sensitivity Index (ESI), the ResXceNet-HBA model achieved 98.5% accuracy, 98.2% precision, 98.7% recall, and 98.4% F1-Score. These results beat ResNet, CNN, and Inception V2. The model executed in 50.9 seconds, faster than the alternatives. The confusion matrix exhibits minimal false positives and negatives, suggesting good prediction accuracy. ResXceNet-HBA s statistics and resource optimization value increases. Precision farming and sustainable agriculture benefit from our strategy s significant environmental stress and crop health assessments.