A decision-fusion strategy for fruit quality inspection using hyperspectral imaging

The focus is on the development of a multi-band decision-fusion strategy for improving the performance of hyperspectral-imaging based fruit inspection systems. The goal is to estimate the ratio of the bruised to unbruised areas in fruits such as strawberries so that the quality can be evaluated by comparing the ratio to a threshold. The challenge, therefore, is to accurately segment the edible part of the berry and accurately classify the pixels in the edible part as either bruised or unbruised. A multi-band segmentation algorithm was formulated to generate a mask for extracting pixels in edible region of a strawberry from each band in a hypercube. A notable feature of the segmentation algorithm is that it is autonomous and does not require prior information to select specific spectral bands in different hypercubes. A multi-band decision-fusion strategy was formulated to exploit information from multiple spectral bands in order to improve the classification of bruised-unbruised pixels. Using a sample of 30 strawberries, experiments were designed to evaluate and compare the performance of the decision-fusion strategy with the best uni-band and multivariate classifiers. It was shown that the improvement in performance using the decision-fusion strategy was statistically significant. Finally, it was noted that the formulation of the multi-band decision-fusion strategy is quite general and is applicable to polychotomous inspection of various other fruits and vegetables.