Ensemble ResNet–XGBoost Framework for Intelligent Crop Disease Detection in Smart Agriculture

Abstract

Crop diseases represent one of the most significant threats to global food security and agricultural sustainability, causing substantial reductions in both yield quantity and produce quality. The timely and accurate identification of plant pathogens remains critically important for modern farming operations, yet conventional detection approaches continue to rely heavily on visual scouting by agricultural experts, a process that proves both time-intensive and economically burdensome, particularly across extensive cultivation areas. This research presents an intelligent crop disease detection system that synergistically combines the representational power of residual neural networks with the optimization capabilities of extreme gradient boosting. The proposed methodology initially applies comprehensive preprocessing procedures including normalization and data augmentation to enhance image quality and expand dataset diversity. Subsequently, the ResNet50 architecture serves as a deep feature extractor, capturing nuanced disease indicators such as chromatic aberrations, necrotic lesions, and textural anomalies from leaf imagery. These extracted high-dimensional features are then fed into an XGBoost classifier, which performs optimized multiclass disease categorization through its ensemble of decision trees. This hybrid configuration demonstrates superior classification accuracy while simultaneously mitigating overfitting concerns and exhibiting enhanced generalization across varying agricultural conditions. The framework effectively discriminates between healthy foliage and multiple disease states, thereby facilitating integration into real-time agricultural monitoring infrastructures. The experimental evaluation confirms that this approach enables early pathogen detection, supports informed decision-making for intervention strategies, and ultimately contributes to reduced crop losses, minimized pesticide utilization, and the advancement of sustainable precision farming methodologies in contemporary agriculture.

Plant phenotyping relevance

葉画像から病徴を抽出し、ResNet50とXGBoostで植物の健全・病害状態を分類する手法が研究の中心であり、評価も実施している。

Code and data availability

The paper uses the public PlantVillage dataset and TensorFlow/Keras/scikit-learn, but provides no authors' public code, model checkpoints, or dataset deposit URL. The data availability statement only says data is included in the submitted report, and no qualifying asset URL appears among the allowed URLs.

No evidence-backed public reproduction asset is currently recorded.