Transfer learning for the semantic segmentation of cryoshpere radargrams

García, Miguel Hoyo; Donini, Elena; Bovolo, Francesca

doi:10.1117/12.2600237

Cited by 5 publications

(6 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In a number of studies radargrams were analysed to find different segments or subsurface targets (e.g. englacial boundaries, EFZ, basal units) and classes of events in each radargram (e.g., Donini et al, 2019;Goldberg et al, 2020;García et al, 2021García et al, , 2023. Even though we focus on the methods for mapping englacial ice structure and tracing IRHs and/or layer boundaries, we also take a look at studies done to detect regions and targets in radar products, since those are, in terms of methodology, in close vicinity to stratigraphy mapping endeavours.…”

Section: Timeline Of the Methods Usedmentioning

confidence: 99%

See 1 more Smart Citation

Review article: Feature tracing in radio-echo sounding products of terrestrial ice sheets and planetary bodies

Moqadam,

Eisen

2024

Preprint

View full text Add to dashboard Cite

Abstract. Radio-echo sounding (RES) is a useful technique for measuring the subsurface properties of ice sheets and glaciers. One of the most important and unique outcomes is the mapping of ice sheets' englacial layer stratigraphy, mainly consisting of isochronous reflection horizons. Mapping those is still a labour-intensive task. This review provides an overview of state-of-the art (semi-)automated methods for identifying ice surface, basal, and internal reflection horizons from radargrams in radioglaciology. We discuss a variety of methods which were developed or applied to RES data over the last decades, including image processing, statistical techniques, and deep learning approaches. For each approach, we briefly summarize their procedures, challenges, and potential applications. Despite major advances, we conclude that gaps remain in effectively mapping internal reflection horizons in an automated way, but with deep learning representing a potential advancement. This paper aims to inform researchers and practitioners in radioglaciology about current and future trends in mapping the englacial stratigraphy of ice sheets.

show abstract

Section: Timeline Of the Methods Usedmentioning

confidence: 99%

“…Overall, they were more successful in detecting the air and noise classes compared to others. Moving away from unsupervised learning, García et al (2021) attempt to reach the same objectives as Garcia et al…”

Section: Applications Of Deep Learning Methodsmentioning

confidence: 99%

Review article: Feature tracing in radio-echo sounding products of terrestrial ice sheets and planetary bodies

Moqadam,

Eisen

2024

Preprint

View full text Add to dashboard Cite

show abstract

“…In order to avoid further reduction of the spatial resolution, parallel dilated convolution is adopted, which is embodied in ASPP structure. Then ASPP with different expansion coefficients can effectively capture multi-scale information, however, in practice, the larger the sampling rate, the less the number of effective filter weights [32]. In order to overcome this problem and integrate global semantic information into algorithm model, this paper uses image level features, with the structure shown in Fig.…”

Section: Networkmentioning

confidence: 99%

Automatic Detection of Basal Units beneath Antarctic Ice Sheet in Radargram Based on Deep Learning

Wang¹,

Xu²,

Lang³

et al. 2023

AETR

View full text Add to dashboard Cite

Sea level rise, caused by accelerated melting of glaciers in Greenland and Antarctica in recent decades, has become a major concern in scientific, environmental, and political arenas. A comprehensive study of subglacial conditions and processes is particularly important for reliable analysis about their future evolution. Basal units – visibly distinct englacial structures near ice-bedrock interface, could provide substantial insight into subglacial processes, ice-sheet dynamic history and dramatically influence the flow of surrounding ice. In order to enable improved characterization of these features, we develop and apply an algorithm that allows for automatic detection of basal units, therefore, an algorithm based on deep learning is proposed. Compared with traditional method based on manual feature extraction, proposed algorithm can achieve completely automatic recognition of basal units in radargram. The network is mainly composed of ResNet and ASPP module, which can achieve high accuracy in very small dataset. We use radar data collected by Polar Geophysics Group (PGG) of The Earth Institute at Columbia University in Antarctica in 2008-2009 and 2009-2010 for experiments, results confirm the effectiveness and robustness of proposed algorithm. At the same time, a more rapid deep learning method is tried, which uses the lightweight network MobileNet V3 as backbone, obtaining a network structure that can save 81.8 percent of training time and 92.2 percent of processing time with high accuracy. It provides a possibility for rapid network training, application on mobile devices and real-time processing of radargram.

show abstract

“…Deep learning has been applied to radargrams for i) detecting the ice stratigraphy, 10,11 ii) simulating RS images with generative models, 12 iii) target detection, 13 and iv) supervised semantic segmentation. [14][15][16] Most proposed methods are supervised and require a large training set of data and ground truth couples. In the case of semantic segmentation, a label is required for each pixel considered in the training phase.…”

Section: Introductionmentioning

confidence: 99%

“…However, a large labeled dataset, which is required to avoid overfitting, is hard to retrieve in the RS domain. Although several methods use mitigation techniques to reduce the number of labeled data for effective training, [14][15][16] there is a need for a method to segment radargrams in an unsupervised way.…”

Section: Introductionmentioning

confidence: 99%

Unsupervised semantic segmentation of radar sounder data using contrastive learning

Donini

Amico²,

Bruzzone³

et al. 2022

Image and Signal Processing for Remote Sensing XXVIII

View full text Add to dashboard Cite

Radar Sounders (RSs) are active sensors widely used for planetary exploration and Earth observation that probe the subsurface in a non-intrusive way by acquiring vertical profiles, called radargrams. Radargrams contain information on subsurface geology and are analyzed with neural networks for segmentation and target detection. However, most of these methods rely on supervised training, which requires a large amount of labeled data that is hard to retrieve. Hence, a need emerges for a novel method for unsupervised radargram segmentation. This paper proposes a novel method for unsupervised radargram segmentation by analyzing semantically meaningful features extracted from a deep network trained with a contrastive logic. First, the network (encoder) is trained using a pretext task to extract meaningful features (query). Considering a dictionary of possible features (keys), the encoder training loss can be defined as a dictionary look-up problem. Each query is matched to a key in a large and consistent dictionary. Although such a dictionary is not available for RS data, it is dynamically computed by extracting meaningful features with another deep network called the momentum encoder. Secondly, deep feature vectors are extracted from the encoder for all radargram pixels. After the feature selection, the feature vectors are binarized. Since pixels of the same class are expected to have similar feature vectors, we compute the similarity between the feature vectors to generate a cluster of pixels for each class. We applied the proposed method to segment radargrams acquired in Greenland by the MCoRDS-3 sensor, achieving good overall accuracy.

show abstract

Transfer learning for the semantic segmentation of cryoshpere radargrams

Cited by 5 publications

References 12 publications

Review article: Feature tracing in radio-echo sounding products of terrestrial ice sheets and planetary bodies

Review article: Feature tracing in radio-echo sounding products of terrestrial ice sheets and planetary bodies

Automatic Detection of Basal Units beneath Antarctic Ice Sheet in Radargram Based on Deep Learning

Unsupervised semantic segmentation of radar sounder data using contrastive learning

Contact Info

Product

Resources

About