Abstract:

A model for citrus weight measuring was developed to realize online grading of citrus fruit with computer vision system. Four images from each of mid-maturing Lianhong Wenzhou citrus fruits (grown in the middle part of Hunan province) captured by computer vision system were processed by cutting, background removal, binary conversion and negative operation and finally the images of fruit part were extracted. The pixel ratios of fruit part/total image were used to construct partial least square (PLS) regression model to estimate fruit weight. Results showed that the absolute weight estimation error was －14.9092 －4.9981 g and the average error was －3.9278 g. The standard deviation was 4.5210 g, and the correctness for accuracy ±10 g was 93.3333% and for accuracy ± 8 g 76.6667%. It is feasible to estimate citrus fruit weight online and monitor the fruit weight during growing period by computer vision system.

Key words: citrus fruit, weight, digital image processing, partial least square