We describe S2LET, a fast and robust implementation of the scale-discretised wavelet transform on the sphere. Wavelets are constructed through a tiling of the harmonic line and can be used to probe spatially localised, scale-depended features of signals on the sphere. The scale-discretised wavelet transform was developed previously and reduces to the needlet transform in the axisymmetric case. The reconstruction of a signal from its wavelets coefficients is made exact here through the use of a sampling theorem on the sphere. Moreover, a multiresolution algorithm is presented to capture all information of each wavelet scale in the minimal number of samples on the sphere. In addition S2LET supports the HEALPix pixelisation scheme, in which case the transform is not exact but nevertheless achieves good numerical accuracy. The core routines of S2LET are written in C and have interfaces in Matlab, IDL and Java. Real signals can be written to and read from FITS files and plotted as Mollweide projections. The S2LET code is made publicly available, is extensively documented, and ships with several examples in the four languages supported. At present the code is restricted to axisymmetric wavelets but will be extended to directional, steerable wavelets in a future release.
A Deep Learning & Fast Wavelet Transform-based Hybrid Approach for Denoising of PPG Signals
这封信提出了一种新的混合方法,该方法利用深度学习来利用小波的多分辨率分析能力,以对光电容积脉搏波(PPG)信号进行降噪。在所提出的方法下,长度为N的噪声PPG序列首先使用快速小波变换(FWT)分解成L个详细系数。然后,如下重建干净的PPG序列。自定义前馈神经网络(FFNN)为逆FWT块输出的每个小波子信号提供二进制权重。这样,在重建期间丢弃所有那些对应于噪声或伪影的子信号。FFNN在监督学习框架下在Beth Israel Deaconess Medical Center(BIDMC)数据集上进行训练,由此我们计算去噪序列与参考干净PPG信号之间的均方误差(MSE),并计算反向传播的MSE梯度。数值结果表明,该方法能有效地去除PPG和视频PPG信号中的噪声。