Shoaib et al. (2026) Plant stress detection using multimodal imaging and machine learning: from leaf spectra to smartphone applications

Identification

• Journal: Frontiers in Plant Science
• Year: 2026
• Date: 2026-01-02
• Authors: Muhammad Shoaib, S. H. Khan, Hala Abdelhameed, Ayman Qahmash
• DOI: 10.3389/fpls.2025.1670593

Research Groups

• Department of Computer Science, CECOS University of IT and Emerging Sciences, Peshawar, Pakistan
• Department of Computer Science and Information Technology, Faculty of Information Technology, The University of Lahore, Lahore, Pakistan
• Informatics and Computer Systems Department, King Khalid University, Abha, Saudi Arabia
• Faculty of Computer and Artificial Intelligence, Fayoum University, Fayoum, Egypt
• Khaybar Applied College, Taibah University, Medina, Saudi Arabia

Short Summary

This review synthesizes advancements in optical sensing technologies and machine learning approaches for detecting biotic and abiotic plant stresses, comparing traditional and modern imaging techniques. It highlights the emerging role of portable and smartphone-based platforms in democratizing access to scalable, automated stress diagnostics for sustainable agriculture.

Objective

• To synthesize recent advancements in optical sensing technologies and machine learning approaches for detecting biotic and abiotic plant stresses.
• To clarify the relative strengths, limitations, and practical applicability of traditional versus modern imaging-based techniques.
• To highlight the emerging role of portable and smartphone-based platforms in democratizing access to stress diagnostics and how machine learning enables scalable, automated analysis.

Study Configuration

• Spatial Scale: Leaf-level to canopy-level, covering greenhouse and field studies, small plots, and large-area monitoring.
• Temporal Scale: Focus on early and pre-symptomatic detection, real-time monitoring, and time-dependent physiological responses.

Methodology and Data

• Models used: Support Vector Machines (SVM), Artificial Neural Networks (ANN), Random Forest (RF), Convolutional Neural Networks (CNN) (e.g., VGG, AlexNet, GoogLeNet, ResNet, U-Net, YOLOv8-Seg, EfficientNet), Transformer models.
• Data sources: Hyperspectral imaging (350–2500 nm), Multispectral imaging/spectroscopy (365–1100 nm), RGB imaging (400–750 nm), Thermal imaging (7.5–14 µm), Fluorescence spectroscopy/imaging (337–760 nm). Data acquired from handheld devices, smartphones, and Unmanned Aerial Vehicles (UAVs).

Main Results

• Multimodal imaging (RGB, NIR, SWIR, PAM fluorescence) combined with machine learning offers significant potential for continuous crop health monitoring and early stress detection.
• Hyperspectral imaging provides the highest sensitivity for detecting subtle biochemical and structural changes but is costly and generates large data volumes.
• Multispectral imaging offers a balance between information content and affordability/portability, suitable for targeted stress indices.
• RGB imaging, especially via smartphones, is highly accessible and low-cost but has limited diagnostic specificity and is sensitive to environmental variations.
• Thermal imaging effectively detects water stress through leaf temperature changes but lacks specificity for different stress types.
• Fluorescence methods are highly sensitive to photosynthetic efficiency perturbations, providing early physiological indicators, but often require controlled conditions.
• Machine learning algorithms (SVM, RF, ANN, CNN, Transformers) automate feature extraction, classification, and prediction, significantly improving the accuracy and scalability of stress detection.
• Deep learning models (CNNs, Transformers) achieve state-of-the-art performance for image-based and multimodal data, especially with large, annotated datasets, while classical ML models are effective for smaller, spectral datasets.
• Multimodal data fusion enhances diagnostic precision and robustness by leveraging complementary information from different sensors.
• Smartphone-based solutions are emerging as cost-effective, field-deployable platforms, democratizing access to plant stress diagnostics.

Contributions

• Provides a comprehensive synthesis of recent advancements in optical sensing and machine learning for plant stress detection.
• Offers a comparative decision matrix (Tables 4a, 4b) to guide the selection of appropriate sensing modalities and machine learning algorithms based on performance, cost, portability, and scalability.
• Clarifies the relative strengths, limitations, and practical applicability of various diagnostic methods, from traditional techniques to modern imaging and smartphone-based solutions.
• Outlines a prioritized 5–10 year research roadmap, including recommendations for standardization, data efficiency, multimodal fusion, explainable AI, and scalable deployment.
• Emphasizes critical considerations for machine learning models, such as data limitations, overfitting, interpretability, and the need for robust validation and benchmarking protocols.

Funding

• King Khalid University, Large Research Project (grant number RGP2/283/46)

Citation

@article{Shoaib2026Plant,
  author = {Shoaib, Muhammad and Khan, S. H. and Abdelhameed, Hala and Qahmash, Ayman},
  title = {Plant stress detection using multimodal imaging and machine learning: from leaf spectra to smartphone applications},
  journal = {Frontiers in Plant Science},
  year = {2026},
  doi = {10.3389/fpls.2025.1670593},
  url = {https://doi.org/10.3389/fpls.2025.1670593}
}