Detection of surface defects on fruits using spherical intensity transformation

The non-uniform intensity distribution on the fruit's images is the main reason resulting in the difficulty and low accuracy of surface defects detection by using a machine vision system. A detection system based on Vis-NIR double CCDs was built for detecting surface defects on 'Akesu' apples. A spherical intensity transformation method (SITM) was proposed to transform the R channel image of an apple, which enhanced the intensity uniformity of the normal regions and kept the low intensity of the defected regions in an apple. The intensity contrast between the defect regions and those of normal tissue was also improved, which increased the defect detection accuracy. A defect detection algorithm was developed based on the SITM and 100 apples consisting of 45 defected apples and 55 intact apples were used to evaluate the performance of the algorithm. Results showed that 93.3% of defected apples were correctly classified and 100% of the intact apples were correctly recognized. The overall detection accuracy was 97%. It is feasible to extract the surface defects on apples using the proposed SITM combining with a single threshold segmentation method.