A Hybrid CNN–Tabular Ensemble Framework for Interpretable Pathogen Classification in Hydroponic Lettuce

Abdulaziz Aborujilah, Ibraheem Abdullah Mohammed Shayea, Waled Hussein Al-Arashi, Samir Hammami, Husam N. Yasin

2026

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.20428/jst.v31i7.3945
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Hydroponic cultivation of lettuce is highly susceptible to bacterial and fungal disease outbreaks, driven by the shared nutrient solution and dense canopy characteristic of controlled-environment agriculture. Early and accurate pathogen identification is therefore essential for crop protection. Deep learning has advanced automated plant disease recognition considerably, but two problems persist for hydroponic lettuce specifically: publicly available datasets remain small and class-imbalanced, and bacterial and fungal symptoms often present overlapping visual signatures that pure CNN representations struggle to separate. This paper proposes a hybrid ensemble framework combining deep CNN embeddings with hand-crafted tabular descriptors—color indices (Excess Green, Color Index of Vegetation Extraction), grey-level co-occurrence matrix texture features, and shape descriptors—into a joint representation that preserves both semantic content and biologically interpretable disease cues. Random Under-Sampling Boosting (RUSBoost) is applied on top of this fused representation as an interpretable, imbalance-aware classifier suitable for deployment on resource-constrained hydroponic edge devices. Evaluated on the public Hydroponic Vegetables-v2 dataset, the framework achieves per-class ROC–AUC scores of 0.878 for bacterial, 0.859 for fungal, and 0.912 for healthy classes, showing that the hybrid representation improves separability of visually overlapping pathogen categories under imbalanced conditions. The contribution lies not in a new architecture but in the specific integration of interpretable descriptors, deep features, and imbalance-aware ensemble learning tailored to the operational constraints of controlled-environment agriculture, offering a reproducible baseline for hydroponic lettuce disease monitoring.