Synergizing End-to-End Spatio-Temporal Feature Extraction with FRPGW-Optimized HALSTM: A Unified Evolutionary Deep Learning Framework for Holistic Precision Agriculture.
Year: 2026
Type: book
Abstract: Precision agriculture has rapidly evolved from simple monitoring and rudimentary sensor deployments to complex, data-driven decision-making systems operating at the edge and cloud levels. However, existing methodologies often treat spatial tasks (such as visual plant disease detection and pest classification) and temporal tasks (such as long-term yield prediction, growth stage monitoring, and micro-climatic forecasting) in absolute isolation. This modal isolation fails to capture the holistic dependencies inherent in complex agro-ecosystems. Objective: This paper proposes a novel, unified, and comprehensive deep learning framework: Synergizing End-to-End Spatio-Temporal Feature Extraction with Fuzzy-Rough Parallel Grey Wolf (FRPGW) Optimized Hierarchical Attention Long Short-Term Memory (HALSTM). The primary goal is to seamlessly bridge the gap between static spatial disease diagnosis and dynamic temporal growth modeling within a single end-to-end differentiable pipeline. Methods: We introduce a sophisticated Hierarchical Attention Long Short-Term Memory (HALSTM) network designed to capture multi-scale temporal phenological dependencies, coupled inextricably with a customized, attention-augmented ResNet-based backbone (SE-ResNet) for high-fidelity spatial feature extraction. To address the profound challenge of hyperparameter selection and architectural tuning in this heavily parameterized hybrid architecture, we develop the Fuzzy-Rough Parallel Grey Wolf Optimizer (FRPGW), a highly advanced bio-inspired evolutionary algorithm. FRPGW incorporates fuzzy-rough set theory for uncertainty-aware initialization and a parallel pack communication strategy to adaptively tune the model and guarantee optimal convergence across a highly non-convex loss landscape. The framework further utilizes a multimodal gating strategy to fuse high-dimensional spatial embeddings with sequential micro-climatic data representations. Results: Extensive and rigorous experiments were conducted on a massive, composite, multi-modal dataset comprising over 50,000 high-resolution hyperspectral and RGB images, temporally aligned with 5 years of continuous micro-climatic sensor data. The proposed FRPGW-HALSTM framework achieved an unprecedented disease classification accuracy of 98.2%, a pest detection precision of 97.8%, and a yield prediction Root Mean Square Error (RMSE) of 0.12. This significantly outperforms an array of state-of-the-art base and hybrid models. Conclusion: The empirical results definitively demonstrate the efficacy of the FRPGW-HALSTM synergy in delivering highly robust, low-latency, and high-precision agricultural diagnostics. This unified, end-to-end approach provides a fundamentally scalable and deeply optimized solution for modern smart farming paradigms, directly addressing critical global challenges in sustainable intensification and food security.
Authors: Anju Markose, K. Baalaji K. Baalaji
Institution: Bharath University
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: gold
OpenAlex ID: W7204875953