Image Classification Using Hybrid Classical-Quantum Neutral Networks

With the increasing complexity of image classification tasks, traditional convolutional neural networks face performance bottlenecks when dealing with intricate network structures. To address this issue, this paper proposes an image classification model based on a hybrid quantum-classical neural network. This model incorporates parameterized quantum circuits into convolutional networks to achieve a hybrid embedding and direct output. The information from fully connected layers is utilized as control parameters for quantum layers, and the pixel values of images are mapped to quantum bit states through amplitude encoding. This allows the network to simultaneously process information from multiple pixels. Furthermore, the introduction of the quantum natural gradient algorithm aims to better handle the geometric properties of the quantum parameter space, accelerating model convergence and improving training efficiency. Experimental results demonstrate an increase in recognition accuracy of \(2.42 \%\) $2.42\%$ and \(5.21 \%\) $5.21\%$ on the MNIST dataset compared to convolutional networks and fully connected networks, respectively. When compared to other algorithms of the same type, the proposed algorithm shows an improvement of \(2.14 \%\) $2.14\%$ and \(3.23 \%\) $3.23\%$ , showcasing superior classification performance.