Research On Optical Device Transmission Signal Recognition Based On Deep Spiking Neural Network

The study of optical device transmission signals is important for many other fields such as sensors.As artificial intelligence evolves,researchers are gradually trying to combine optical device transmission signals with deep learning.However,optical device transmission signals are not easily collected in large quantities and are mostly noisy,which leads to traditional Artificial Neural Networks(ANN)that do not achieve very satisfactory results in recognition tasks.In recent years,it has been pointed out that Spiking Neural Network(SNN)can surpass ANN in the study of "small sample with noise" problems,so SNNs are considered to be applied to the study of optical device transmission signals.The main research of this paper is as follows.1.Since the transmission signals of optical devices are not easily collected in large quantities and are mostly noisy,we consider the transmission signals of optical devices as a "small sample with noise" problem.Therefore,we verify that SNNs are more suitable for studying "small sample with noise" problems based on public datasets,and then illustrate the feasibility of studying optical device transmission signal identification based on deep Spiking Neural Networks.2.To address the problem that SNNs are mostly still limited to shallow structures with low recognition accuracy,we constructed a Residual Attention Spiking Neural Network(RASNN)by adding two modules,residual structure,and ECA-Net,to the fully connected SNN.Then we conducted comparison experiments on our established one-dimensional photonic crystal transmission signal dataset and demonstrated that the proposed RASNN can achieve better recognition results.Finally,we also conducted comparison experiments with the ANN commonly used in existing studies based on the MIT-BIH arrhythmia dataset to demonstrate that RASNN can also provide a robust solution to the recognition problem of other one-dimensional signals that are not easily collected in large quantities.3.The metasurface transmission signal dataset is established,and a Temporal Convolutional Spiking Neural Network(TCSNN)is constructed by adding two modules of the temporal convolutional module and residual structure to the fully connected SNN.The comparison experiments based on the metasurface transmission signal dataset are conducted to evaluate the network performance.Finally,the recognition effects of RASNN and TCSNN are compared on the one-dimensional photonic crystal transmission signal dataset and the metasurface transmission signal dataset,respectively.It is shown that the two proposed deep Spiking Neural Network models are comparable in recognition,and both can accurately and efficiently identify the transmitted signals of optical devices.