Ran Dong

In the hilly and mountainous areas of Japan, mowing operations can only be carried out by human labor because of the steep slopes. However, the environment faced by workers when mowing is complex, requiring them to deal with different visual stimuli at the same time. These factors will also be reflected in the data of specific pupil changes, further impacting their concentration while mowing. Therefore, in this study, based on a set of experiments on various terrain (flat land and slope) in Hiroshima, Japan, an analysis method of human pupil changes was proposed based on action decomposition technology Hilbert–Huang Transform (HHT) which can be used to calculate the different frequency patterns (intrinsic mode function, IMF) that represent nonlinearity in pupil changes more effectively than Fourier Transform or Wavelet Transform. Based on the use of our proposed Multiple Comparisons and Filtering framework named MCFID, the IMFs which directly related to specific mowing actions (cutting and lifting) were found though the statistical tools. By monitoring the corresponding IMFs, it is possible to calculate the period of the corresponding pupil movement, and further inversely infer information such as the subject’s concentration status. Our approach can also be validated using other pupil movement datasets. The results of the study can provide useful insights for training new lawn mowers, and the relevant data can be used as data accumulation for the development of future fall detection systems.

In this study, the Hilbert-Huang Transform (HHT) and other frequency analysis methods, Short-Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT), are employed to numerically analyze electromagnetic phenomena with strong non-linearity. In the numerical experiments, electromagnetic wave propagation simulations using the Finite-Difference Time-Domain (FDTD) method are performed to reproduce the behavior of electromagnetic shielding effects leaking from the shielding plate. The obtained results were analyzed using the HHT, STFT, and CWT methods; the HHT analysis showed that electromagnetic waves around the input wavelength 2.45 GHz leak significantly from the shielding plate. On the other hand, the STFT did not accurately resolve the signal, and the CWT unexpectedly showed strong amplitudes in the low-frequency region. These results suggest that HHT performs better in analyzing signals with strong non-linearity and provides more accurate results than other methods.

Multivariate Empirical Mode Decomposition (MEMD) is a powerful tool to analyze nonlinear properties, however, it is very computationally time-consuming. The purpose of this study is to parallelize and accelerate MEMD, and to apply the accelerated MEMD to the analysis of electromagnetic wave propagation in photonic crystals, to improve the efficiency of generation of a demultiplexer. In the proposed method, part of the procedures that do not depend on channels of input signal were parallelized. Speedup which is based on 24 CPU threads of 5.12 times is achieved when 192 threads are adopted.

In this study, a novel pre-processing method to improve the prediction accuracy of deep learning (DL) in predicting motor characteristics are proposed. In the proposed method, a frequency decomposition method called empirical mode decomposition is applied to the gap magnetic flux density distribution, which is the input data of the DL.

In the Finite-Difference Time-Domain (FDTD) method, the analysis domain is divided into orthogonal meshes, the space is discretized by the finite difference method, and the time step is advanced by the Leap-Frog method. Therefore, it is difficult to accurately model analysis regions with complex shapes or smooth curved surfaces, such as modeling a helical filter with spiral curves to generate an optical vortex. When the filter is modeled with orthogonal meshes, it results in a staircase-like representation. For this reason, reflections at the gaps on the staircase surface will occur in the computational results as non-physical phenomena, which require sufficient numerical verification. In this study, the HE ₁₁ mode is injected into a corrugated waveguide, and a helical filter is set to generate an optical vortex using the FDTD method. Moreover, the Empirical Mode Decomposition (EMD) is adopted to remove gap reflections and extract the clear physical phenomenon of the optical vortex. The results revealed that our method could extract the generated optical vortex correctly by removing the non-physical phenomena that occurred in the electric field as a post-processing technique. In addition, these results have also been quantitatively verified in the time-frequency domain using spectrum analysis.

In this paper, we propose an improved version of the Pyramidal Predictive Network (PPNV2), a theoretical framework inspired by predictive coding, which addresses the limitations of its predecessor (PPNV1) in the task of future perception prediction. While PPNV1 employed a temporal pyramid architecture and demonstrated promising results, its innate signal processing led to aliasing in the prediction, restricting its application in robotic navigation. We analyze the signal dissemination and characteristic artifacts of PPNV1 and introduce architectural enhancements and training strategies to mitigate these issues. The improved architecture focuses on optimizing information dissemination and reducing aliasing in neural networks. We redesign the downsampling and upsampling components to enable the network to construct images more effectively from low-frequency-input Fourier features, replacing the simple concatenation of different inputs in the previous version. Furthermore, we refine the training strategies to alleviate input inconsistency during training and testing phases. The enhanced model exhibits increased interpretability, stronger prediction accuracy, and improved quality of predictions. The proposed PPNV2 offers a more robust and efficient approach to future video-frame prediction, overcoming the limitations of its predecessor and expanding its potential applications in various robotic domains, including pedestrian prediction, vehicle prediction, and navigation.