Data fusion of two hyperspectral imaging systems for blueberry bruising detection

Abstract. A push broom based hyperspectral imaging (HSI) system in the spectral region of 700–960 nm and a liquid crystal tunable filter (LCTF) based HSI system in the spectral region of 960–1650 nm, were used to detect blueberry bruising 30 min to 12 h after mechanical impact. A total of 704 blueberry samples from 2 varieties were divided into four groups: one control group, and three bruise treatment groups with bruises created at stem, equator, and calyx positions, respectively. Mean reflectance spectrum of each sample was extracted from push broom based and LCTF based HSI. Combined with partial least squares-discriminant analysis (PLS-DA), the spectral data from the two spectroscopic techniques were analyzed separately, and then the spectral data were fused in three data fusion strategies (data level, feature level and decision level). Classification results showed that the LCTF based HSI outperformed push broom based HSI in blueberry bruising detection. The three data fusion strategies achieved better classification results than using push broom based or LCTF based HSI individually. The decision level data fusion based on weighted majority vote obtained more promising results, demonstrating that the information obtained from the two spectroscopic techniques could make a synergistic effect on improving blueberry internal bruising detection.