The Digital Eye in Plant Disease Surveillance: A Critical Synthesis of Imaging, Artificial Intelligence and Smart Diagnosis
Pooja Naduvalath *
Department of Plant Pathology, College of Agriculture Padannakkad, Kerala Agricultural University, Kerala, India.
P. Sainamole Kurian
Department of Plant Pathology, College of Agriculture Padannakkad, Kerala Agricultural University, Kerala, India.
P. K. Sajeesh
KVK, Palakkad, India.
I. L. Shahana
Department of Computer Science, College of Agriculture, Padannakkad, India.
P. R. Aswathi
Department of Plant Pathology, College of Agriculture Padannakkad, Kerala Agricultural University, Kerala, India.
Athira Suresh
Department of Plant Pathology, College of Agriculture Padannakkad, Kerala Agricultural University, Kerala, India.
P. T. Sneha
College of Agriculture, Padannakad, Kerala Agricultural University, Kerala, India.
Aksa Anna Shajan
Department of Plant Pathology, College of Agriculture, Vellanikkara, Kerala Agricultural University, Kerala, India.
*Author to whom correspondence should be addressed.
Abstract
Plant disease surveillance is moving from episodic visual inspection towards image-supported observation across leaves, crop canopies and landscapes. Digital cameras, spectral and thermal sensors, uncrewed aerial vehicles and artificial intelligence have expanded the measurable manifestations of plant stress, yet the ability to recognise a pattern is not equivalent to identifying its cause or improving disease management. This critical narrative review examines the evidence linking imaging modalities, computational models and diagnostic workflows to operational surveillance. Peer-reviewed studies and methodological reviews published from 2010 to 10 September 2026 were considered, with earlier concepts addressed where necessary. The synthesis evaluates sensor-specific biological signals, experimental design, dataset provenance, cross-domain generalisation, object detection and severity estimation, aerial and satellite monitoring, and the role of human and laboratory verification. Controlled-image classification has demonstrated substantial technical feasibility, but independent field evaluations frequently reveal reductions in recall and weak transportability across sites, cultivars, seasons and acquisition devices. Spectral and thermal measurements can detect altered physiology before obvious visual injury in selected experiments; nevertheless, such responses are seldom pathogen-specific and may be confounded by abiotic stress or crop development. Multi-scale mapping can locate suspected disease foci, although spatial resolution, timing, sampling design and ground-truth uncertainty constrain interpretation. Recent advances in domain adaptation, multimodal learning, anomaly detection and explainability are promising but have not consistently established prospective diagnostic or management value. The review proposes a surveillance-oriented framework that separates detection, aetiological diagnosis, severity estimation, forecasting and intervention. Progress depends less on marginal improvements in benchmark accuracy than on independent prospective evaluation, calibrated uncertainty, representative datasets, expert-mediated confirmation and measurements of decisions and outcomes. Digital imaging is best understood as an additional evidential layer within plant health surveillance rather than a stand-alone substitute for plant pathology.
Keywords: Plant pathology, remote sensing, hyperspectral imaging, domain shift, disease severity, digital epidemiology, diagnostic validation, precision crop protection