OPTIMASI VARIATIONAL QUANTUM CLASSIFIER (VQC) PADA KLASIFIKASI GAMBAR PERKEBUNAN KELAPA SAWIT

Abstract

This study explores the optimization of the Variational Quantum Classifier (VQC) for classifying plantation images, particularly distinguishing between oil palm and non-oil palm trees. Previous quantum models such as Quantum Convolutional Neural Networks (QCNN) demonstrated high accuracy but suffered from high complexity and qubit requirements. To address this limitation, this research employs a more modular and efficient quantum model, the VQC, optimized through the exploration of 18 combinations of feature maps (PauliFeatureMap, ZFeatureMap, ZZFeatureMap), ansatz circuits (RealAmplitudes, TwoLocal, EfficientSU2), and optimizers (COBYLA, SPSA). The dataset comprises 220 images resized to 32×32 pixels and preprocessed through standard scaling, PCA (to 4 dimensions), and normalization to [-1,1]. The best performance was achieved using the combination of ZZFeatureMap, EfficientSU2, and COBYLA, yielding an accuracy of 93%. Evaluation metrics include accuracy, precision, recall, F1-score, and specificity. The results reveal that the selection and interaction of feature map, ansatz, and optimizer significantly affect classification performance. These findings suggest that VQC is a promising alternative for image classification in digital agriculture, offering high accuracy with lower quantum resource demands.