Non-Invasive Detection and Characterization of Powdery Mildew in Strawberries Using Hyperspectral Imaging and Deep Learning under Poly-tunnel Conditions

Powdery Mildew (Podosphaera aphanis, PM) presents a serious challenge to strawberry cultivation, causing significant yield losses if not controlled early. Globally, PM contributes to substantial reductions in strawberry production, posing economic threats to growers and raising food security concerns. Therefore, the development of accurate, noninvasive detection mechanisms is crucial for effective disease management. This study investigates the application of Hyperspectral Imaging (HSI) for nondestructive PM detection and classification of healthy and PM-infected strawberry leaves grown under a poly-tunnel, an environment that closely simulates real-world agricultural conditions. A hyperspectral camera (350-1000 nm), mounted on a mechanized rail system beneath the poly-tunnel, was used to capture leaf images as it moved linearly above the strawberry canopy. Spectral preprocessing included Savitzky-Golay smoothing (SGS), Standard Normal Variate (SNV), and Multiplicative Scatter Correction (MSC), with the Isolation Forest algorithm applied for outlier removal. A one-dimensional Convolutional Neural Network (1D-CNN) was trained to classify healthy versus PM-infected leaves, achieving 75% accuracy and 84% precision. The model outperformed traditional classifiers such as Random Forest (RF), Decision Tree (DT), and Partial Least Squares Discriminant Analysis (PLS-DA). Among all tested pipelines, the SGS+MSC+1D-CNN combination yielded the highest performance. This study highlights the feasibility and effectiveness of integrating HSI with deep learning for robust disease detection under semi-controlled conditions, laying the groundwork for scalable, real-time plant health monitoring in precision agriculture.