Semantic Segmentation for Multi-Contour Estimation in Maritime Scenes

Abstract

Figure 1: A 360 • image overlaid with the extracted true horizon line (red), visible horizon line (green) and shoreline (blue).

Automatic Shoreline Detection by Processing Planview Timex Images Using Bi-Lstm Networks

“…Most of these contributions work with satellite images classifying water-land regions (Li et al, 2018;Shamsolmoali et al, 2019;Liu et al, 2020;Cui et al, 2020;Dang et al, 2022;Seale et al, 2022) by using the CNNs: UNet, SegNet, DeepLabV3+, and variations of them. The sea-land segmentation through CNNs is also applied to images captured by cameras with autonomous navigation purposes (Yao et al, 2021;Finlinson & Moschoyiannis, 2022). We highlight that Liu et al (2020) extracts the contour directly by performing a gradient on the segmented image, an idea that Finlinson & Moschoyiannis (2022) type that can learn dependencies between samples separated in different ranges along with the sequence index, overcoming a limitation exhibited by the original RNNs, which only captured dependencies between nearby samples (Hochreiter & Schmidhuber, 1997).…”

“…The sea-land segmentation through CNNs is also applied to images captured by cameras with autonomous navigation purposes (Yao et al, 2021;Finlinson & Moschoyiannis, 2022). We highlight that Liu et al (2020) extracts the contour directly by performing a gradient on the segmented image, an idea that Finlinson & Moschoyiannis (2022) type that can learn dependencies between samples separated in different ranges along with the sequence index, overcoming a limitation exhibited by the original RNNs, which only captured dependencies between nearby samples (Hochreiter & Schmidhuber, 1997). In practice, simple RNNs have a limited ability to learn long-term dependencies.…”