International Conference on Soft Computing for Security Applications (ICSCSA)
A Hybrid Quantum–Classical Neural Network for Chilli Leaf Disease Classification: A Variational Quantum Feature-Fusion Approach
J. Senthilkumar, Arjun S, Loshitha K, Dhanasri R, Mohammed Faizal
Abstract
Chili (Capsicum annuum) is an economically important crop that is frequently affected by foliar diseases such as leaf curl, leaf spot, whitefly infestation, and yellowing, which cause substantial yield losses if not detected early. Classical convolutional neural networks (CNNs) have been widely applied to automated leaf-disease recognition, but they demand large labelled datasets, are computationally expensive, and their purely classical feature space limits discrimination between visually similar disease symptoms. This paper proposes a Hybrid Quantum-Classical Neural Network (HQ-CNN) that couples a lightweight classical convolutional feature extractor with a parameterised variational quantum circuit (VQC) acting as a trainable quantum feature map. RGB leaf images are pre-processed and passed through convolutional blocks to obtain a low-dimensional embedding, which is angle-encoded into a 4-8 qubit circuit; entangling variational layers followed by Pauli-Z measurement produce quantum features that are fed to a final softmax classifier. The complete network is trained end-to-end using the Adam optimiser for classical weights and the parameter-shift rule for quantum weights, and is evaluated on a five-class, publicly available chilli leaf disease dataset (Healthy, Leaf Curl, Leaf Spot, Whitefly, Yellowish). On this dataset, the proposed HQ-CNN attains 99.1test accuracy with only 24 quantum parameters and a total trainable-parameter count two to three orders of magnitude smaller than classical transfer-learning baselines (traditional CNN, VGG16, MobileNetV2, EfficientNetB0) reported in the literature, and an ablation study confirms that both the presence and the depth of the quantum layer contribute to this result. The results indicate that quantum feature fusion is a promising low-parameter-count direction for fine-grained agricultural image classification, particularly under limited-data regimes typical of plant pathology datasets.