Fig. 4.

Comparison to state-of-the-art segmentation approaches. Four representative field rice images are selected to illustrate the segmentation effect. The first column reflects the original top-view rice images in the field. The second column is the manual panicle segmentation result using Photoshop software. The third column to the sixth column represents the rice panicle segmentation results using HSeg, i2 hysteresis thresholding, jointSeg, and Panicle-SEG algorithm. A The upright panicles are partially hidden in the rice leaf blade. B The bend growth panicles are basically exposed above the rice leaf blade. C The awn exists in the rice panicle, and the illumination is uneven in the same field plot rice image. D On a cloudy day, the panicle color appears to be gray