Comparative Analysis Between Classical CNNs and Quantum Convolutional Neural Networks for Cacao Pod Disease Classification

Abstract

Convolutional Neural Networks (CNNs) have demonstrated high performance in image classification tasks. However, the emergence of Quantum Machine Learning has spurred the exploration of hybrid quantum–classical architectures as a computational alternative. This paper presents a controlled comparative study between a classical CNN and a Quantum Convolutional Neural Network (QCNN) based on amplitude embedding for the binary classification of cacao pod images. A set of 1,000 grayscale images resized to $32\times 32$ pixels was used, employing noiseless quantum simulation with the PennyLane library under a controlled experimental setting inspired by NISQ-era constraints. Different depths of the quantum circuit and their impact on metrics such as accuracy, precision, sensitivity, and F1 score were evaluated. The results show that, although the classical CNN achieves slightly higher predictive performance, the QCNN attains competitive results using a significantly smaller number of trainable parameters. These findings suggest that hybrid quantum–classical architectures can provide compact and parameter-efficient representations for image classification tasks within the scope of noiseless simulation and parameter-constrained experimental settings.