Architectural Design of Quantum Convolutional Neural Network for Multi-Class Image Classification

This study proposes a quantum convolutional neural network (QCNN) designed for image multi-classification tasks. To enhance feature extraction, the quantum convolutional employs singular value decomposition (SVD) to construct custom quantum gates. Quantum fully connected layers are built based on the swap test, which improves both generalization and classification performance. The QCNN is trained and tested on the PennyLane quantum computing framework. It achieves a test accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 4. 0 7 \%}$</tex> on a four-class classification task using the MNIST dataset. To evaluate its practicality, simulation experiments are conducted under various types of quantum noise. The results show that the QCNN remains robust against noise, demonstrating strong potential for practical quantum environments.