Intelligent information extraction from reflectance spectra: Absorption band positions

摘要: A multiple high-order derivative analysis algorithm has been developed that automatically extracts absorption band positions from reflectance spectra. Absorption occur where the fifth of spectrum equals zero, fourth positive sign, and second is negative. The assumes bands are approximately symmetric about center. Continuum contributions, phase angle effects, broad low-frequency calibration errors suppressed. Overlapping with centers as close 0.3–0.5W (full width at half maximum intensity) can be resolved, long have comparable widths intensities. If overlapping dissimilar, center separations 0.6–1.0W safer limits resolution. Results relatively insensitive to whether constituent convolve additively or multiplicatively. Spectral resolution moderately low, requiring only four eight data points per W. Errors derived <3%W for greater than 0.6–1.0W. For a few thousand cm−1 would typically less 150 actual positions. detection sensitive noise, smoothing required. segment length (number averaged) needs continually adjusted ∼0.5W minimize signal distortion. spectral pattern recognition algorithm, which statistically characterizes frequency distribution intensity variations in sliding across spectrum, used predetect (low-frequency components distributions) calculate approximate W (predetected W) using its derivative. An intelligent control then continuously locally adjust lengths 0.5W W). Smooths repeated (typically, 20–30 times) until high-frequency variation distributions single-pass cubic spline applied smoothed data. applies algorithm. sixth-order polynomial fit being 1.0W spectrum. Adjustment ∼1.0W insures minimally distorted weak features not Derivatives calculated point coefficients polynomial. system successfully extracted low-quality (6% peak-to-peak noise) synthetic spectra little degradation accuracy. Application natural laboratory earth-based telescope displayed good reliability consistency. Processing fully automated, same standardized procedure followed all No continuum removal modeling needed. automation could potentially significantly increase efficiency yield information extraction, particularly high-rate repetitive scan synoptic remote sensing spectroscopy applications.