Scanline intersection similarity: A similarity metric for joint trace maps

Although various automatic joint trace survey methods have been proposed, it has been difficult to evaluate their accuracy. This difficulty lies in the lack of comprehensive metrics for expressing the similarity between the automatically derived trace maps and those derived manually. In this study,...

Celý popis

Uložené v:

Podrobná bibliografia
Vydané v:	Computers & geosciences Ročník 175; s. 105358
Hlavní autori:	Kim, Jineon, Lee, Yong-Ki, Choi, Chae-Soon, Fereshtenejad, Sayedalireza, Song, Jae-Joon
Médium:	Journal Article
Jazyk:	English
Vydavateľské údaje:	Elsevier Ltd 01.06.2023
Predmet:	Accuracy evaluation computer vision computers Deep learning-based algorithms Rock discontinuity survey Scanline intersection similarity (SIS) Similarity metric surveys testing Trace map Deep learning-based algorithms Rock discontinuity survey Similarity metric Scanline intersection similarity (SIS) Trace map Accuracy evaluation
ISSN:	0098-3004, 1873-7803
On-line prístup:	Získať plný text
Tagy:	Pridať tag Žiadne tagy, Buďte prvý, kto otaguje tento záznam!

Popis
Shrnutí:	Although various automatic joint trace survey methods have been proposed, it has been difficult to evaluate their accuracy. This difficulty lies in the lack of comprehensive metrics for expressing the similarity between the automatically derived trace maps and those derived manually. In this study, a novel metric is proposed to express the similarity between joint trace maps using a single numerical value. The proposed metric, scanline intersection similarity (SIS), expresses trace map similarity by evaluating the similarity between trace-scanline intersections on many orthogonal scanline pairs. The expressiveness of SIS was tested on multiple synthetic trace map pairs, and the results confirmed that SIS comprehensively reflects trace map similarity in terms of trace frequency, trace orientation, and the spatial distribution of traces. Comparisons with conventional metrics used in computer vision, namely, intersection over union (IoU) and boundary F1 (BF) scores, revealed that SIS expresses trace map similarity more accurately. IoU and BF scores led to misleading conclusions because of their evaluation of trace map similarity using the areal overlap of traces. To further demonstrate the applicability of SIS, the accuracy of deep-learning-based trace detection using U-Net (University of Freiburg, Germany) and DeepLabV3+ (Google, USA) was evaluated and compared using SIS. •A similarity metric for discontinuity trace maps is proposed based on trace-scanline intersections.•The metric comprehensively reflects trace map geometry.•The metric provides a more reliable evaluation of trace map similarity than conventional metrics based on areal overlap.•The applicability of the proposed metric is demonstrated by evaluating the accuracy of deep learning-based trace detection.
Bibliografia:	ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 23
ISSN:	0098-3004 1873-7803
DOI:	10.1016/j.cageo.2023.105358