Semantic segmentation of plant roots from RGB (mini-) rhizotron images—generalisation potential and false positives of established methods and advanced deep-learning models

Table 4 Imaging processing techniques/models tested or developed in this work, source of the methodology and application to root segmentation tasks

Technique/Model	Encoder	Method source	Adapted to root segmentation	Augmented data
Dummy classifier	–	[60]	–	No
Frangi Vesselness	–	[61]	[32]	No
Adaptive thresholding	–	[29]	[28]	No
Support Vector Machine (SVM)	–	[35]	[34]	No
SegRoot	–	[38]	[38]	Yes
UNetGNRes	Default U-Net	[44]	[20, 33]	Yes
U-Net	SE-ResNeXt-101 (32 × 4d)	[62]	This paper	Yes
U-Net	EfficientNet-b6	[62]	This paper	Yes

Encoders are only present in U-Net models; augmented data has been added for model training only

ISSN: 1746-4811