Image segmentation algorithm for greenhouse cucumber canopy under various natural lighting conditions

In this study, machine vision technology was used to capture images of greenhouse cucumber canopy, and image segmentation was implemented under various natural lighting conditions. The images were enhanced by multi-scale retinex with color restore (MSRCR), and the MSRCR images were segmented by four algorithms: normalized difference index (NDI), excess green (ExG), modified excess green (MExG), and modified excess green minus excess red (MExG-ExR). The results indicated that compared with the original images, under various lighting conditions, the average evaluation indexes of brightness, information entropy, average gradient and mean gray value of the MSRCR images were increased by 38.71%, 8.04%, 4.54%, and 37.81%, respectively, and only the contrast degree decreased by 12.13%. The MExG-ExR segmentation algorithm was used to segment the MSRCR images (fifty images under various lighting conditions in the test and it performed best among the four segmentation algorithms, average overlap ratios and recognition rates of were 99.28% and 98.91%, respectively, while 38.39% and 37.95% respectively for original image. These results indicated that the MExG-ExR segmentation algorithm applied to a MSRCR canopy image produced the most stable results among the four algorithms. By using the MSRCR image enhancement algorithm, the interference of lighting on greenhouse cucumber canopy images was reduced and the foundation for achieving accurate segmentation of a canopy region was laid, which is of great significance for greenhouse cucumber phenotypic parameter measurements.