Fractional transform assisted image recognition with genetic diagnostic calibration for plant disease detection

Early and accurate detection of plant diseases is essential to ensure agricultural productivity, minimize crop losses, and maintain food security. Conventional image-based recognition systems, though promising, often face limitations when deployed in real-world farming environments. Variations in lighting, overlapping disease symptoms, and differences between laboratory-curated datasets and field images reduce their robustness and reliability. At the same time, genetic diagnostic tools such as CRISPR-based assays have demonstrated great potential for molecular-level confirmation, yet these remain separated from image recognition workflows, thereby limiting their practicality in field applications. This research introduces FRaGILE, a novel Fractional-transform-augmented Genetic-diagnostics-Integrated Learning framework. The framework uniquely integrates fractional spectral–spatial transforms with deep learning architectures to enhance disease classification accuracy. Fractional Fourier and wavelet transforms are applied to capture lesion texture and frequency features at varying fractional orders, while convolutional and transformer backbones are trained to learn discriminative and invariant representations. Additionally, a Bayesian calibration layer incorporates optional CRISPR-based diagnostic signals, improving confidence estimation and reducing false positives in ambiguous cases. Experimental validation is conducted using multi-source plant pathology datasets, including hyperspectral and visible light benchmarks. Results indicate consistent improvements in F1-score and AUROC compared to baseline convolutional networks. Sensitivity analyses reveal that fractional transform orders within the range of 0.6 to 0.9 yield optimal performance, while the incorporation of fractional Laplacian enhancements further strengthens detection under variable illumination. The proposed system is implemented on a hardware–software stack suitable for smartphones and edge devices, enabling field-level deployment. The hybrid workflow, which begins with image-based triage and leverages genetic confirmation when necessary, establishes a practical, scalable, and scientifically rigorous approach to plant disease detection. By fusing phenotypic and genotypic insights, FRaGILE sets the foundation for more reliable, actionable, and sustainable disease management in agriculture. This study focuses on the challenge of accurate plant disease detection by integrating fractional-transform-based image recognition with genetically informed diagnostic calibration. Conventional image-only approaches often work poorly with early-stage infections and visually ambiguous symptoms due to various factors such as lighting variability, background clutter, and inter-class similarity. In an attempt to bridge this gap, we introduce FRaGILE: Fractional Representation and Genetic Integrated Learning Engine, a hybrid framework that synergistically fuses the phenotypic image features with simulated CRISPR-based molecular evidence. The central research questions which this study tried to answer are: RQ1: Does the fractional-order transform improve robustness of features against noise and variation in illumination? (RQ2) Does calibration of genetic diagnostic improve the confidence and reliability in classification? The proposed hybrid framework can perform better than the state-of-the-art deep learning models on different plant disease datasets.