Research On Image Classification Based On Hybrid Quantum Classical Model

One of the hotspots in quantum computing has always been quantum machine learning,quantum machine learning outperforms classical machine learning in terms of model complexity and performance due to the benefits of entanglement,state superposition,and computational parallelism.The goal of quantum machine learning is to utilize the advantages of quantum computing to improve classical machine learning further,which means that the quantum computing can significantly enhance the developments of machine learning.As quantum machine learning develops,researchers gradually combine the benefits of quantum and classical computing to create a hybrid quantum-classical neural network model,which is consistent with the running ability of quantum computers at the present stage.To enhance the performance of the classical neural network model,this research investigates the construction of hybrid quantum-classical neural network and its performance in a noisy environment based on parameterized quantum circuits and classical neural networks.The main research is as follows:1.A hybrid quantum-classical convolutional neural network model for classical image classification is proposed.The model is composed of quantum state coding,quantum convolutional layer,quantum pooling layer,and classical fully connected layer.The quantum state coding part first completes the mapping of the classical information to the original information quantum state,and then the unitary transformation of the original information quantum state is finished by the quantum convolution kernel in the quantum convolution layer,and the feature information quantum state is obtained.Then the quantum pooling layer is used to map the feature information to reduce the dimension of the feature information quantum state.Finally,the feature information in the quantum state is obtained by quantum measurement,and the feature information is input into the classical fully connected layer for further processing.The model realizes the quantum convolution of the whole image based on the classical convolution position without moving the convolution window and repeating the quantum state coding and measurement operations.The quantum pooling layer is added to the model to replace the classical pooling layer,which reduces the dimension of the quantum state and the measurement times of the quantum system.The simulation results show that the accuracy of the proposed model is better than the classical CNN in image classification,and the effects of the number of quantum convolution kernels and quantum pooling layer on the performance of the model are discussed in the experiment.2.To explore the performance of hybrid quantum-classical neural network model in the noisy environment,a hybrid quantum-classical tensor neural network model is constructed,and its performance under different noise disturbances is discussed.Considering two kinds of noise disturbances: quantum circuit noise and disturbed sample noise.First,under the quantum circuit noise,different noise types and intensities are set for experimental simulation,and a quantum classifier is set up for comparative experiments;in the disturbance sample noise,based on the transferability of disturbance samples in adversarial machine learning,the disturbance samples generated by a similar classical model are used to analyze the hybrid model’s performance and compared with the classical model and quantum model.The experimental results show that,compared with the quantum classifier,the hybrid quantum-classical tensor network model has a stronger ability to resist quantum circuit noise.The simulation results of disturbance sample noise show that the transfer between the classical model and the hybrid quantum-classical model has certain advantages compared with the disturbance sample transfer between the two classical models,but it also proves that the hybrid quantum-classical model is vulnerable to adversarial attack.At the same time,this paper also discusses the performance of the classical model,the hybrid quantum-classical model,and the quantum model in the disturbance samples.The results show that the hybrid quantum-classical model has the best performance among the three models.