量子衍生神经网络模型及其在图像恢复中的应用

Quantum-derived neural network model and its application to image restoration

为提高图像恢复质量,提出一种量子衍生神经网络模型及算法.该模型为3层结构,隐层为量子神经元,输出层为普通神经元.量子神经元由量子旋转门和多位受控非门组成,利用多位受控非门中目标量子位的输出向输入端的反馈,实现对输入序列的整体记忆,利用受控非门输出中多位量子比特的纠缠,获得量子神经元的输出.基于量子计算理论设计了该模型的学习算法,该模型可从宽度和深度两方面获取输入序列的特征.仿真结果表明,该模型的图像恢复效果明显优于普通神经网络.

In order to improve the quality of image restoration, a kind of quantum-derived neural network model and algorithm are proposed. The model is composed of three layers of structure, in which, the hidden layer is made up of quantum neurons, and the output layer is made up of common neurons. The quantum neuron consists of a quantum rotation gate and a multi-qubit controlled not-gate. By using the information feedback of the target qubit from the output to the input end in the muhi-qubit controlled not-gate, the integral memory of input sequences is realized. The output of the quantum neuron is obtained by the entanglement of the muhi-qubit in the output of the con- trolled not-gates. On the basis of the theory on quantum computation, the learning algorithm for the model is designed. Through the model, the characteristics of the input sequence may be effectively obtained from two aspects including ＂width＂ and ＂depth＂. The simulation results show that the quality of image restoration of the model is obviously superior to that of the common neural network.