The effects of changing laboratory illumination zenith angle on spectral feature parameters and mineral classification

A continuously changing solar position affects outdoor hyperspectral measurements and classifications results. It is important to know how much spectra vary with a changing zenith angle and which classification technique is least affected. In this research, we measure spectra from seven pure minerals under nine zenith angles (5 degrees to 85 degrees in steps of 10 degrees) using an Analytical Spectral Devices (ASD) Fieldspec (R) 3 Spectroradiometer in a dark room. The ASD was calibrated using a Spectralon (R) white reference panel every time we adjusted the zenith angle. We created a data-set consisting of the pure mineral spectra and linear mixtures derived from the pure mineral spectra. Spectral variation of the data-set with increasing zenith angle was analyzed through spectral feature parameters. Subsequently, three classifiers: spectral angle mapper (SAM), Euclidean distance classifier (EDC), and an expert system decision tree (ESDT) were used to classify the data-set. Reference spectra used in the SAM and EDC classifications are the pure mineral spectra measured under a zenith angle of 5 degrees. We designed the ESDT using center wavelength position and absorption depth derived from the deepest absorption feature of the pure mineral spectra measured under a zenith angle of 5 degrees. We show that, with increasing zenith angle, the center position of spectral features is quite consistent, while the spectral reflectance is affected, and the continuum removal extends the effect. When classifying pure mineral spectra, SAM and ESDT display robustness of 100% with the changing zenith angles, while EDC displays robustness of 95.24% which is a result of misclassifying three unknown spectra measured under a zenith angle of 85 degrees. The performance of classifiers becomes complicated when classifying the linearly mixed spectra. SAM performs a robustness of 100%, 98.41%, 77.78%, and 33.33% with a spectral mixing fraction of 95%, 85%, 75%, and 50% respectively. EDC performs a robustness of 93.65%, 93.65%, 93.65%, and 36.51% with a spectral mixing fraction of 95%, 85%, 75%, and 50% respectively. ESDT performs a robustness of 100%, 100%, 100%, and 52.38% with a spectral mixing fraction of 95%, 85%, 75%, and 50% respectively. We conclude that the changing zenith angle affects mineral spectra through increasing reflectance values and affecting depth of absorption features. Regarding the mineral classification, the spectral mixture and selection of classifiers present more effects than the changing zenith angle.