Automated online detection of granulation in oranges using X-ray radiographs

Oranges can develop granulation during production, a condition in which the juice sacs shrivel because of gel formation. The aim of this work is to develop an image processing algorithm to reliably detect granulation in X-ray projection images or radiographs of oranges. Oranges grown at South-African orchards with a known high incidence of granulation were scanned in an X-ray system (75 kV, 468 mA, 60-ms exposure). Subsequently they were destructively evaluated on the presence of granulation to serve as a ground truth. An image processing algorithm was developed to automatically segment the affected fruit tissue and train a naive Bayes classifier based on the spatially discretized features. The resulting high-speed and robust algorithm can be implemented in existing on-line X-ray radiograph systems. When applied in a sorting line, this should result in sampling ratios close to 100%, with no losses of healthy fruit due to destructive testing. Furthermore the end product will have a guaranteed low presence of granulation, increasing commercial value.