【摘要】 提出了一种基于卷积神经网络并用于估算受像散透镜影响的涡旋光束拓扑荷数的方法。将携带不同拓扑荷数的涡旋光束经过不同像散程度的透镜所产生的衍射光斑图片作为数据集，分别输入到VGG、AlexNet、ResNet-18、ResNet-34、ResNet-50和Xception等6种经典的神经网络模型中，采用特异度、精确率、召回率、F1分数以及准确率作为衡量网络模型估算结果的评估函数，分别对整张光斑图片以及部分光斑图片的拓扑荷数进行估算。结果表明，采用所提方法对拓扑荷数的估算都取得了良好效果。相比于其他网络模型，VGG网络模型在此估算任务上表现更好，对拓扑荷数的估算准确率达到了99%以上。这表明使用所提方法估算经过像散透镜后涡旋光束所携带的不同拓扑荷数是可行的，该方法可进一步应用在光束通过光学系统传输的特征识别领域。更多还原

【Abstract】 In this study, a method based on the convolutional neural network is developed to estimate the topological charge of vortex beams affected by astigmatic lenses. Vortex beams with different topological charges are incident on lenses with different astigmatism coefficients. The generated diffraction intensity images are input into six classical neural network models, which are VGG, AlexNet, ResNet-18, ResNet-34, ResNet-50, and Xception. Specificity, precision,recall, F1-score, and accuracy are used as evaluation functions to measure the topological charge estimated by the network model. The network models are used to estimate the topological charge of the entire and part intensity images. The results show that the proposed method is effective for estimating the topological charge. The VGG model exhibits the best performance in this task, and the accuracy of topological charge estimation is more than 99%. Therefore, it is feasible to use neural network to estimate the different topological charges carried by the vortex beams after passing through the lens with astigmatism. This method can be further applied in the feature recognition of beam transmission through optical systems.更多还原