基于DenseNet的经典-量子混合分类模型

Hybrid classical-quantum classification model based on DenseNet

现有的图像分类模型越来越复杂,计算时所需的硬件资源和计算时间不断增加。针对该问题提出一种基于DenseNet的经典-量子混合分类模型(CQDenseNet模型)。首先,使用一个可在噪声中尺度量子(NISQ)设备上运行的变分量子电路(VQC)作为分类器,替换DenseNet全连接层;其次,使用迁移学习,利用在ImageNet数据集上预先训练好的DenseNet模型作为CQDenseNet的预训练模型;最后,将CQDenseNet模型在中草药分类数据集和CIFAR-100数据集上与基准模型AlexNet、GoogLeNet、VGG19、ResNet和DenseNet-169进行对比。实验结果表明,CQDenseNet模型比所有基准模型中表现最好的基准模型:准确率分别提高了2.2、7.4个百分点,精确率分别提高了2.2、7.3个百分点,召回率分别提高了2.2、7.1个百分点,F1值分别提高了2.3、6.4个百分点,说明了经典-量子混合模型的性能优于经典模型。

Existing image classification models are becoming more and more complex,and the hardware resources and computation time required for computation are increasing.A hybrid Classical-Quantum classification model based on DenseNet(CQDenseNet model)was proposed to address this problem.First,a Variational Quantum Circuit(VQC)that could operate on a Noisy Intermediate-Scale Quantum(NISQ)device was used as a classifier to replace the fully connected layer of DenseNet.Secondly,by using transfer learning,a pre-trained DenseNet model on the ImageNet dataset was utilized as a pre-training model for CQDenseNet.Finally,CQDenseNet model was compared with the benchmark models AlexNet,GoogLeNet,VGG19,ResNet and DenseNet-169 on Chinese Medicine and CIFAR-100 datasets.Experimental results show that CQDenseNet model is more effective than the best-performing benchmark model,with improvements of 2.2 and 7.4 percentage points in accuracy,2.2 and 7.3 percentage points in precision,2.2 and 7.1 percentage points in recall,and 2.3 and 6.4 percentage points in F1-score,respectively.It shows that the performance of the hybrid classical-quantum model is better than the classical models.