Table 3.

Segmentation accuracy of the thresholding methods (HSV only, LAB only, and combined HSV and LAB) and the U-net method and finally the difference between HSV/LAB and U-net.

	HSV	LAB	HSV/LAB	U-net	Diff
Jaccard index (IOU)	0.65	0.75	0.77	0.84	+0.07
Sørensen-Dice coefficient	0.76	0.85	0.86	0.91	+0.05
Pixel accuracy	0.89	0.93	0.93	0.95	+0.02
Precision	0.66	0.80	0.83	0.89	+0.06
Recall	0.97	0.93	0.91	0.93	+0.02
AP [0.5 IoU]	0.91	0.93	0.94	0.96	+0.02
F1 score	0.78	0.86	0.87	0.91	+0.04

Segmentation accuracy measures, Jaccard index or intersection over union (IoU), the Sørensen-Dice coefficient, pixel accuracy, precision, recall, F1 score estimated from the precision, and recall and AP for predictions over 0.5 IoU as described by the PASCAL VOC challenge (Everingham et al., 2010).