Ex Vivo Assessment of Bacterial Heart Rot Symptoms in ‘MD2’ Pineapple Leaves Using VIS-NIR Spectroscopy and Machine Learning
Аннотация
Bacterial heart rot (BHR), primarily associated with the pathogen Dickeya zeae , represents a critical constraint to ‘MD2’ pineapple production in tropical regions, with documented field losses reaching up to 40% in Malaysia and 50% in Indonesia. Traditional diagnostic protocols currently rely on visual scouting and confirmatory laboratory procedures that are inherently labor-intensive, subjective, and difficult to scale across expansive commercial plantations. To address these limitations, this preliminary proof-of-concept study evaluates an ex vivo screening framework by integrating visible to near-infrared (VIS-NIR) reflectance spectroscopy (420–920 nm) with machine learning analysis on detached ‘MD2’ pineapple leaves. To increase disease-related signals, 310 leaf samples from four plantation sites in Peninsular Malaysia were analyzed using standardized preprocessing, including Savitzky–Golay smoothing and standard normal variate transformation. Using a plant-grouped validation split to prevent data leakage, partial least square regression demonstrated predictive capability for relative chlorophyll status (SPAD, $$R^{2}=0.79$$) and objective CIELAB color attributes. A sensitivity-prioritized random forest classifier achieved a test recall of 0.918, albeit accompanied by a low specificity of 0.154 due to class imbalance and conservative thresholding. Furthermore, a simplified two-band spectral disease index (SDI at 466.592 nm and 663.573 nm) demonstrated limited test discrimination (area under the curve [AUC] = 0.553), whereas the photochemical reflectance index (PRI) achieved higher overall diagnostic accuracy (AUC = 0.724). These findings establish an ex vivo baseline under laboratory darkroom conditions, highlighting both the potential and current limitations of optical spectral profiling prior to intact, open-field validation.