Formal Verification and Deep Learning for Smart Farming: A Robust Framework for Cotton Crop Monitoring

Abstract

Cotton, a critical global cash crop, faces significant challenges in disease detection due to wetland conditions and climate-change inconsistency. This work presents a Cotton Crop Disease Detection Model that integrates an EfficientNet--Convolutional Neural Network (CNN) architecture with the Temporal Logic of Actions (TLA+) for formal verification. The proposed model ensures accurate disease classification while providing formal verification for correctness, reliability, and availability. The EfficientNet--CNN demonstrates robust performance in identifying multiple disease conditions, including aphids, armyworms, and bacterial blight, achieving an overall weighted accuracy of 94%, with macro-average scores of 0.94 for precision, recall, and F1-score. Class-specific performance shows an F1-score of 97% for armyworms and 96% for powdery mildew. The TLA+ formal verification validates the model's compliance with disease-monitoring requirements, ensuring correctness, reliability, and availability in real-world industrial applications. This integrated framework enhances cotton crop disease detection and supports sustainable, technology-driven agricultural practices.

Plant phenotyping relevance

綿花の病害状態を画像ベースのCNNで分類する手法を開発・検証しており、植物の状態推定が中心的な技術貢献である。

Code and data availability

The supplied blocks describe an EfficientNet-CNN cotton disease detection framework with a TLA+ formal model, but contain no public dataset link, code repository, trained model release, or availability statement. The cotton leaf dataset is described only narratively (2137 original / 7000 augmented images), and the TLA+

No evidence-backed public reproduction asset is currently recorded.