Robust materials classification based on multispectral polarimetric BRDF imagery

When light is reflected from object surface, its spectral characteristics will be affected by surface's elemental composition, while its polarimetric characteristics will be determined by the surface's orientation, roughness and conductance. Multispectral polarimetric imaging technique records both the spectral and polarimetric characteristics of the light, and adds dimensions to the spatial intensity typically acquired and it also could provide unique and discriminatory information which may argument material classification techniques. But for the sake of non-Lambert of object surface, the spectral and polarimetric characteristics will change along with the illumination angle and observation angle. If BRDF is ignored during the material classification, misclassification is inevitable. To get a feature that is robust material classification to non-Lambert surface, a new classification methods based on multispectral polarimetric BRDF characteristics is proposed in this paper. Support Vector Machine method is adopted to classify targets in clutter grass environments. The train sets are obtained in the sunny, while the test sets are got from three different weather and detected conditions, at last the classification results based on multispectral polarimetric BRDF features are compared with other two results based on spectral information, and multispectral polarimetric information under sunny, cloudy and dark conditions respectively. The experimental results present that the method based on multispectral polarimetric BRDF features performs the most robust, and the classification precision also surpasses the other two. When imaging objects under the dark weather, it's difficult to distinguish different materials using spectral features as the grays between backgrounds and targets in each different wavelength would be very close, but the method proposed in this paper would efficiently solve this problem.