Scalable quantum convolutional neural network for image classification

Quantum machine learning (QML) is a promising area of research that combines the capability of quantum computing with machine learning approaches with the goal of outperforming traditional computers while processing vast amounts of data and solving challenging problems. Meanwhile, Convolutional Neural Networks (CNNs) excel in areas such as image classification, by extracting features more efficiently than traditional neural networks. Although traditional CNNs have shown good performance in image classification, training CNNs demands significant computational resources. Quantum machine learning also confronts several obstacles and constraints at present, and the number of qubits as well as the scalability of quantum computers need to be enhanced. In this paper, we propose a technique termed Scalable Quantum Convolutional Neural Networks (SQCNN) to overcome these limitations. We implemented the model on the TensorFlow Quantum platform, where we carried out simulations with the MNIST and Fashion MNIST datasets. The experimental results reveal that SQCNN achieves an average classification accuracy of 99.79%. Compared with existing quantum neural network models, our model not only has higher classification accuracy, but also demonstrates strong performance across other evaluation metrics. It is worth mentioning that the quantum circuit we designed draws on the idea of convolutional neural networks, which can better learn features by relying on superposition and entanglement between quantum gates. In particular, multiple independent quantum devices in the SQCNN system can extract features in parallel. This design allows the flexible use of quantum devices of different sizes, thereby achieving larger-scale machine learning tasks.