ntroduction – Using machine vision (MV) systems for automatic grading of fruit usually imposes high costs on horticulture industry and needs complicated programing for image processing. The purpose of this study was, thus, to grade citrus fruit using the mass and volume estimation from their projected area (PA). The results can be used to design simple and cheap one-camera MV systems for automatic grading of citrus fruit quality. Materials and methods – The actual mass, volume, and PA of 100 fresh citrus fruit (including oranges, tangerines and grapefruits) were measured by using a digital scale, water displacement method and an area meter, respectively. The adaptive neuro-fuzzy inference system (ANFIS) and Gaussian process regression (GPR) models were also used to estimate the mass and volume of fruit from their PA. In turn, grid partition (GP), sub-clustering (SC) and fuzzy c-means (FCM) methods were used for fuzzy inference system (FIS) building. Moreover, using the mass and volume estimator models in line with the median value, the fruit were sorted in two small and big classes. Results and discussion – In the FCM-ANFIS method considered as the best model, the mean absolute percentage error (MAPE) values of the mass and volume estimation were found to be 1.88% and 2.56% for oranges, 2.73% and 2.76% for tangerines, and 3.06% and 2.73% for grapefruits, respectively. Besides, the classification accuracy values for the mass of oranges, tangerines and grapefruits were 93%, 88% and 87%, while for the fruit volume they were found to be 93%, 90% and 91%, respectively. Conclusion – The comparison of ANFIS and GPR showed that ANFIS had better performance and less error in grading the fruit quality. Furthermore, the FCM type ANFIS model was identified an appropriate method for the mass and volume estimation of citrus fruit using their PA

Keywords