Non-Destructive Detection of Plant Leaf Diseases Using Digital Image Analysis

Early detection of plant diseases is essential for improving crop productivity and supporting global food security. Conventional diagnostic methods are often destructive, labor-intensive, and unsuitable for continuous monitoring. In this context, digital image-based approaches have emerged as a key component of non-destructive testing (NDT) in agriculture. This paper presents a comprehensive review of image processing and computer vision techniques for leaf disease detection, explicitly framed within NDT applications. The literature is systematically organized using a dual taxonomy based on feature representation (color, texture, morphological patterns) and learning paradigm (traditional machine learning versus deep learning), enabling a direct comparison of methodological capabilities and limitations. The analysis reveals that color-based methods are efficient but unreliable under variable illumination, while texture descriptors improve robustness in detecting structurally complex symptoms such as necrosis. Deep learning approaches, particularly convolutional neural networks, achieve high accuracy in controlled conditions (frequently above 90%) but exhibit performance degradation in field environments, with reported drops linked to background variability, occlusion, and inconsistent lighting. Across studies, a clear trade-off emerges between model accuracy and deployment feasibility, with high-performing models requiring large annotated datasets and significant computational resources. The review further identifies a lack of standardized evaluation protocols and limited cross-dataset validation as major barriers to practical NDT deployment. Although low-cost imaging platforms such as smartphone-based systems enable scalable data acquisition, their integration into reliable real-time diagnostic pipelines remains insufficiently addressed. In conclusion, this comprehensive review not only synthesizes existing methods but also highlights critical gaps that hinder the transition from controlled experiments to real-world NDT applications, particularly in terms of robustness, generalization, and system-level integration within precision agriculture.