Table 3.

Evaluation of the mask R-CNN models in low-density validation dataset using different DA techniques for the detection of hachure mound features: random translation (DA1), random rotation (DA2) and the so-called Doppelgänger technique (DA3).

Algorithm	TPs	FNs	FPs	Recall (%)	Precision (%)	F1 (%)
None	87	21	737	80.56	10.56	18.67
DA1	71	39	37	64.55	65.74	65.14
DA1 + DA2	68	45	53	60.18	56.20	58.12
DA1 + DA2 + DA3	68	44	31	60.71	68.69	64.45