Multisensor X-ray inspection of internal defects in horticultural products

A combination of 3-D vision and X-ray radiography is proposed to enable low-cost, generally applicable online inspection of internal quality of horticultural and potentially other products. The underlying concept assumes that the shape of the product is known beforehand through a deformable shape model. A 3-D vision system is used in combination with the shape model to accurately determine the complete outer surface shape of the sample. This shape is voxelized to generate a reference product from which a X-ray radiograph is simulated to be compared with a measured radiograph, hence revealing the presence of any defects or disorders. Advantages of this method are that small deviations in internal density are detected easily since the cumulative information of the bulk object shape is removed. Furthermore, no specific detection algorithms have to be developed for different types of defects, since the method will directly identify deviations from the ideal. Validation on two datasets and comparison with two reference detections methods (classical image processing and a human operator) shows that the proposed method reliably (accuracy >99%) detect defects larger than 3.5 mm radius with densities differences between sample and defects as small as 10%. Voids are reliably (accuracy >99%) detected down to a radius of 1.5 mm, corresponding to a volume of less than 0.02 cm3.