Hartley transform

About: Hartley transform is a research topic. Over the lifetime, 2709 publications have been published within this topic receiving 79944 citations.

Comparison of feature extractors on DC power system faults for improving ANN fault diagnosis accuracy

Abstract: The power system operator's need for a reliable power delivery system calls for a real-time or near-real-time AI-based fault diagnosis tool. These needs are universal, whether they be for terrestrial-based or nonterrestrial-based power delivery systems, namely the NASA Space Station Alpha (Alpha). In this paper, we present a comparison of feature extractors suitable to the training and consultation phases for a fault diagnosis tool based on a two-stage ANN clustering algorithm. One of the prime concerns in selecting an appropriate feature extractor is to provide the ANN with enough significant details in the pattern set so that the highest degree of accuracy in the ANN's performance can be obtained. Candidate feature extractors include time domain analysis, frequency-domain analysis using the fast Fourier transform and the Hartley transform, and wavelet domain analysis using the wavelet transform. Simulated fault studies on a small system are performed and results presented to illustrate the performance capabilities of the respective feature extractor coupled ANN clustering algorithm sets.

Faster image super-resolution by improved frequency-domain neural networks

Abstract: Recently, most deep learning-based studies have focused on elaborately developing various types of neural networks in the spatial domain to tackle super-resolution. These methods usually have numerous parameters and require a huge amount of memory and time to train their networks. A frequency-domain neural network for super-resolution (FNNSR) has been presented. It designs a primitive neural network that can be trained using back propagation in the frequency domain. In this paper, we propose an improved FNNSR. In our method, the parameters of four quadrants in the compact weighting layer are shared. This substantially reduces the number of parameters in this layer. We use multiple convolutional layers with activations instead of a single convolution operator in FNNSR, so that the underlying features in transformed images can be learned. The Hartley transform is computed directly rather than through the Fourier transform. We design a new weighted Euclidean loss, which pays more attention to reconstructing the high-frequency part that is difficult to recover. Extensive experiments show that our method runs faster and requires fewer parameters than FNNSR. Though our method is not better than state-of-the-art methods in terms of quantitative performance, it still reveals encouraging results.

Generalized fourier-feynman transform and sequential transforms on function space

Abstract: In this paper werst investigate the existence of the gener- alized Fourier-Feynman transform of the functional F given by F ( x) = ^ (( e1;x) � ;:::; ( en;x) � ) ; where ( e;x) � denotes the Paley-Wiener-Zyg stochastic integral with x in a very general function space Ca;b(0 ;T ) and ^ is the Fourier transform of complex measure on B( R n ) withnite total variation. We then dene two sequential transforms. Finally, we establish that the one is to identify the generalized Fourier-Feynman transform and the another transform acts like an inverse generalized Fourier-Feynman transform.