12 August 2020 in IEEE Access
Due to the lack of data available for training, deep learning hardly performed well in the field of garbage image classification. We choose the TrashNet data set which is widely used in the field of garbage image classification, and try to overcome data deficiencies in this field by optimizing the network structure. In this paper, it is found that the deeper network and short-circuit connection, which are generally accepted in the field of deep learning, will not work well on the TrashNet data set. By analyzing and modifying the network structure, we propose an effective method to improve the network performance on TrashNet data set. This method widens the network by expanding branches, and then uses add layers to realize the fusion of feature information. It can make full use of feature information at slight additional computational cost. Using this method to replace the core structure of the Xception network, the performance of the improved network has been improved greatly. Finally, the M-b Xception network proposed by us achieves 94.34% classification accuracy on the TrashNet data set, and has certain advantages over some state-of-the-art methods on multiple indicators. The python code can be download from https://github.com/scp19801980/Trash-classify-M_b-Xception.
1 July 2026 in IEEE Photonics Technology Letters
The enhanced Gaussian noise (EGN) model, which accounts for inter-channel stimulated Raman scattering (ISRS), enables accurate evaluation of nonlinear interference (NLI) across the C+L band. However, its computation time increases linearly with the number of wavelength channels under test, limiting its practical applicability. This letter proposes a channel NLI extrapolation method that requires computation on only five sparsely selected wavelength channels, enabling real-time, integration-free predictions across the entire transmission bandwidth. The proposed method accommodates arbitrary launch power distributions and does not rely on large-scale data training. Over a 20×100 km transmission distance, the mean absolute error (MAE) of SNRNLI below 0.1 dB, with the maximum error for any channel not exceeding 0.3 dB across various modulation formats, spectral configurations, fiber types, and launch powers. This approach enhances the engineering applicability of the ISRS EGN model and offers a computationally efficient framework for accurate NLI modeling.
30 June 2025 in IEEE Photonics Journal
This study proposes a learned digital backward propagation (LDBP) algorithm that performs a distributed compensation of polarization mode dispersion (PMD) in parallel. The proposed LDBP algorithm uses the regular perturbation theory of a fiber-optic nonlinear Schrödinger equation to create a deep neural network (DNN) with full parallelization capabilities. The proposed algorithm’s nonlinear compensation (NLC) performance is evaluated using numerical experiments on a 1,000-kilometer standard single-mode fiber link. The link uses a dense wavelength division multiplexing (DWDM) system with five wavelength channels, 64-QAM modulation, and a symbol rate of 32 GBaud/s per channel. The experimental results demonstrate that, even for low-complexity network training with identical optical power, the proposed method can provide a performance improvement of approximately 0.4 dB over lumped compensation methods. This indicates that parallelization, which improves the efficiency of NLC execution, does not reduce the proposed method’s advantage over lumped compensation methods. Finally, the improvement of the computational efficiency caused by parallelization is investigated. The results show that parallelization improves computational efficiency by about 67 times compared to serial execution. The findings of this paper provide a feasible solution for implementing NLC, which can significantly improve hardware efficiency in practice.
6 April 2026 in Optics Express
The enhanced Gaussian noise (EGN) model, which incorporates inter-channel stimulated Raman scattering (ISRS), offers a highly accurate and comprehensive physics-based framework for evaluating nonlinear interference (NLI) in optical fiber communication systems. However, its practical applicability is limited by its high computational cost. In this work, we identify the oscillatory behavior of the integrand in the four-wave mixing (FWM) efficiency factor as a key bottleneck and propose an approximation method that yields an accurate closed-form expression for the FWM efficiency factor. Numerical results demonstrate that the proposed method achieves a low mean absolute error (MAE) of approximately 0.0033 dB under strong ISRS conditions, while reducing computation time by more than 99.6% compared to the full integral evaluation. Additionally, we develop and optimize a parallel computation strategy that exploits independent regions of the integration domain, further enhancing overall computational efficiency.
1 November 2026 in IEEE Photonics Technology Letters