DisQuan: a hybrid quantum–classical architecture for plant disease detection in staple and specialty crops

Abstract

Image-based plant disease identification is essential for advancing smart agriculture, particularly for staple crops such as rice and economically significant crops such as tea, which are cultivated under complex environmental conditions. Furthermore, these crop groups present distinct challenges: rice leaf disease data typically exhibits clear pathological structures but necessitates large-scale deployment on resource-limited devices, whereas tea leaf disease data is complicated by variable lighting, diverse backgrounds, and high biodiversity. Although current deep learning models achieve high accuracy, they predominantly utilize deep convolutional neural network (CNN) architectures with millions of parameters, which hinders practical deployment on edge devices and increases the risk of overfitting when field data is scarce. In response, this study introduces DisQuan, a hybrid architecture that integrates classical deep learning with quantum machine learning (QML) to balance accuracy and resource efficiency. In particular, DisQuan combines the lightweight DisNet feature-extraction network with a variablequantum neural network to compress and refine feature representations in quantum space, yielding a model with only 0.09 million parameters. Experimental results on rice and tea leaf disease datasets indicate that DisQuan achieves the highest accuracy on the rice dataset and performance comparable to deep CNN models with significantly more parameters on the tea dataset, while maintaining a compact and stable structure. Overall, these findings suggest that DisQuan provides a practical compromise between performance and model complexity, and highlight the potential of quantum-classical hybrid architectures for plant disease detection in real-world agricultural settings and on resource-constrained devices.

Plant phenotyping relevance

植物葉の病害状態を画像から識別する量子・古典ハイブリッド手法を開発し、複数データセットで性能とモデル規模を評価しており、表現型取得・推定が中心である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.