SmartRoot-AI: A Multimodal Deep Learning Framework for Plant Stress Detection and Root Architecture Analysis

Precision agriculture demands timely and accurate identification of plant stress to optimize crop productivity and resource management. Traditional visual inspection methods are subjective, labor-intensive, and fail to assess critical underground root health indicators. This paper presents SmartRoot-AI, a comprehensive multimodal agricultural intelligence system integrating ensemble deep learning, computer vision, sensor fusion, and explainable artificial intelligence (XAI) for holistic plant and root health analysis. The system employs an ensemble of convolutional neural networks (CNNs) including custom lightweight architectures, ResNet50, and EfficientNetV2-S for robust plant stress detection. Advanced root architecture analysis utilizes U-Net for semantic segmentation and RootNav-inspired structural extraction techniques. Environmental parameters including temperature, humidity, and soil moisture are fused with visual predictions through a confidence-weighted fusion mechanism, enhancing system robustness under variable field conditions. Gradient-weighted Class Activation Mapping (Grad-CAM) provides model interpretability by highlighting discriminative image regions. A Streamlit-based interactive dashboard facilitates real-time visualization and supports offline deployment in resource-constrained environments. Experimental evaluation demonstrates that ensemble inference achieves superior reliability and reduced prediction variance compared to single-model approaches, while the multimodal fusion strategy effectively mitigates false positives under challenging environmental conditions.