Neural networks for pattern recognition
Neural networks for pattern recognition
Learning associative memory predictors from time series
Signal Processing
Data Mining and Knowledge Discovery: Making Sense Out of Data
IEEE Expert: Intelligent Systems and Their Applications
| Hi-index | 0.00 |
Recent advances in machine learning methods, and their successful applications across a multitude of fields ranging from astronomy to bioinformatics, offer a promise of powerful new tools for scientists. This paper proposes a novel hybrid neural network architecture and learning method for some tasks in spectrum analysis, namely identification of spectral bands and peaks, and spectrum fitting. The proposal uses a combination of multilayer feedforward networks (FFNs) and Radial Basis Function (RBF) networks. Starting from a mathematical model of a spectrum as a sum of overlapping curves representing spectral peaks, we train a FFN to detect peak positions and half-widths. Using these parameters, we obtain the remaining spectrum model parameters (the intensities I) from spectral data. The resulting RBF network is then used to fit the measured spectrum. Experiments with synthetic and measured spectra demonstrate the capability of the proposed technique for construction of new intelligent systems for spectrum analysis and estimation.