CITRUS FRUIT IDENTIFICATION USING MACHINE VISION FOR A CANOPY SHAKE AND CATCH HARVESTER
A machine vision-based citrus yield mapping system was developed for a canopy shake and catch harvester. The system consisted of aluminum housing, a 3CCD progressive scan digital color camera, and four halogen lamps for illumination. The housing was built to hold the lamps and the camera for image acquisition, and also to keep the sunlight from going in the conveyor belt when images are acquired. With the system, images were acquired from a testing bench as well as from a com¬mercial canopy shake and catch harvester. The images were divided into calibration and validation sets. For accurate yield estimation, an algorithm was developed to identify the citrus fruit in the images. The colors of fruit and background pixels in the calibration images were investigated in different color spaces (RGB, HSI, YIQ, and YCbCr) and threshold values were determined to separate fruit from background. Simple image processing steps were implemented to obtain a clean binary image. A marker-controlled watershed method was used to separate clusters of fruit. The number of fruit, area, and diameter were calculated and compared with the actual number of fruit. The R2 value between the actual number of fruit and the number of fruit counted by the algorithm was 0.956 for the test bench testing and 0.642 for the harvester testing. When fruit diameter measured from calibration images was used, the R2 value between the actual number of fruit and the number of fruit estimated by the fruit diameter was 0.874 for the test bench testing and 0.906 for the harvester testing.
Won Suk Lee, , Chinchuluun, R. and Ehsani, R. (2009). CITRUS FRUIT IDENTIFICATION USING MACHINE VISION FOR A CANOPY SHAKE AND CATCH HARVESTER. Acta Hortic. 824, 217-222
citrus, identification, machine vision, yield mapping