由于大气湍流的影响,涡旋光在大气中传播时会产生相位畸变,造成接收端模态检测困难,导致通信系统的可靠性降低.为提高涡旋光束模态检测准确率,提出了利用简单无参注意力卷积神经网络(S-Conv NeXt)检测拉盖尔-高斯光束模态的算法.结果发现:该网络可有效聚焦于关键亮斑特征.当光束分别经过较弱湍流、中等湍流、较强湍流和强湍流传输 2km后,本征态检测准确率分别为100%、98.8%、96.4%和 89.7%,叠加态检测准确率分别为 100%、99.8%、98.8%和96.5%.在强湍流下,S-ConvNeXt 网络本征态检测准确率比ResNet50、ShuffleNetV2、Conv-NeXt网络分别提高5.7%、3%、1.2%,叠加态检测准确率分别提高 5.7%、4%、0.9%.S-Conv-NeXt网络能够有效提高模态检测准确率,尤其在强湍流条件下表现更好.
When vortex beam propagates in the atmosphere,phase distortion is generated due to the influence of atmospheric turbulence,which makes it difficult to detect the mode at the receiving end and reduces the reliability of the communication system.In order to improve the accuracy of vortex beam mode recognition,a simple parameter-free attention convolution neural network(S-ConvNeXt)is proposed.Results show that this proposed network can effectively focus on key bright spot features.When the transmission distance is 2 km,the accuracy of eigenstate recognition can reach 100%,98.8%,96.4%,89.7%,the accuracy of superposition state recognition can reach 100%,99.8%,98.8%,96.5%,via weak turbulence,medium turbulence,strong turbulence and stronger turbulence respectively.Under strong turbulence,the eigenstate recognition accuracy of S-ConvNeXt is 5.7%,3%and 1.2%higher than that of ResNet50,ShuffleNetV2 and Conv NeXt,and the superposition state recognition accuracy of S-Conv NeXt is 5.7%,4%and 0.9%higher than that of ResNet50,ShuffleNetV2 and ConvNeXt respectively.S-Conv NeXt can effectively improve the accuracy of mode recognition,especially in strong turbulence.
魏冬梅;刘芳宁;杜乾;王珂;赵曰峰
山东师范大学物理与电子科学学院光场调控及应用中心,山东济南 250358
电子信息工程
涡旋光束;Conv NeXt网络;大气湍流;模态识别;注意力机制
vortex beam;ConvNeXt;atmospheric turbulence;mode recognition;attention mech-anisms
《陕西师范大学学报（自然科学版）》 2024 (002)
111-120 / 10
国家自然科学基金(42271093);山东省本科教学改革研究项目(M2021235)
10.15983/j.cnki.jsnu.2024309
评论