1. Introduction
In recent years, global warming has exacerbated the melting of glaciers in Greenland and Antarctica, which has considerable influence on sea level rise and, subsequently, the safety of those living in coastal and seaside locations. In order to gather data about glacier melting, it is vital to identify and quantify changes in the ice and bedrock layers of these glaciers. Historically, glaciologists probed the subsurface structure of ice sheets in polar regions by drilling ice cores, but newer methods—such as ground-penetrating radar (GPR) technology—enable scientists to gather robust data sets quickly and efficiently.
Prior work on this topic involves analyzing data from radar sounder instruments (known as radargrams or echograms) to draw inferences about the properties of the ice sampled. Radar sounders, which are usually operated on airborne or satellite platforms, are active instruments that can perform non-intrusive depth measurements of the subsurface structure of the ice sheets on a large spatial scale. While the advent of ground-penetrating radar made the data collection process more efficient, the data analysis process is still intensely time-consuming because it is typically done by hand.
Newer work in this field utilizes image processing, computer vision, and deep learning techniques to automatically or semi-automatically determine ice surface and bottom boundaries from echograms [
1,
2,
3,
4,
5,
6,
7]. Gifford et al. [
1] employs both the edge-based and active contour methodologies to automate the task of locating polar ice and bedrock layers from airborne radar data acquired over Greenland and Antarctica. In the edge-based approach, edge detection, thresholding, and edge following are utilized to identify the layers of interest for the ice thickness estimation. The active contour approach involves fitting a contour to the boundary using image and contour costs, as well as a gradient force that pushes the contour upward from the bottom of the image. Both methods have their pros and cons; the edge-based method is more efficient, but the active contour method produces more robust results [
1]. A third technique, the level-set model, is better at identifying curved boundaries by embedding the evolving curve into a higher dimensional surface. Mitchell et al. [
2] uses a level-set technique for estimating bedrock and surface layers, but find it problematic because of the need to reinitialize the curve manually for each radar image. Therefore, Rahnemoonfar et al. [
3] introduces the distance regularization term, which essentially maintains the regularity of the level-set and leads to a stable numerical procedure without the need for reinitialization. Recently, Rahnemoonfar et al. [
4] proposes a novel approach that automatically detects the complex topology of the ice surface and bottom boundaries based on the charged particle concept. This method detects the contours in the image based on Coulomb’s Law and the assumption that each pixel is an electrically charged particle. Another approach is to use graphical models to detect ice layers in radar echograms. Lee et al. [
5] frame boundary tracking as an inference problem and employs a Markov Chain Monte Carlo (MCMC) technique to sample from the joint distribution over all possible layers that exist on a given image. Xu et al. [
6] revisits this issue and uses a tree-reweighted message passing (TRW) technique, which first generates a seed surface subject with a set of constraints, and then incorporates additional sources of evidence to refine it via discrete energy minimization. Berger et al. [
7] further improves upon the method of Xu et al. [
6] by incorporating additional domain-specific knowledge into the cost functions and modeling algorithms.
There are also some efforts to achieve the automatic classification of ice subsurface targets in echograms [
8,
9,
10,
11]. Ilisei et al. [
8,
9] exploits the statistical properties of the radar signal to pre-process the radar image data in order to generate a statistical map of the subsurface, and then deploys a segmentation algorithm that had been attuned to that specific study area. Bruzzone et al. [
10,
11] develops an automatic classification system for subsurface targets, which includes extraction of relevant features based on both the statistical properties of the radar sounder signals and the spatial distribution of the ice subsurface targets. Once specific features are identified, they are then categorized with a support vector machine (SVM) classifier. The classification of each category can be used to focus on the study of specific areas. The classification of the whole bedrock area can be used for geological studies, to understand the thickness of bedrock, to evaluate the type of bedrock material and to derive the absorption characteristics of bedrock [
12]. Meanwhile, the thickness of the ice column can be calculated by the classification of layers and bedrock.
At present, the Center of Remote Sensing of Ice Sheets (CReSIS) provide a wide variety of radar data from various radar equipments. A radargram provided by CReSIS) is a 2-D matrix with nR rows (samples i) and nC columns (frames j). The typical radargram model includes free space, layers, echo-free zone (EFZ), bedrock, and noise, as shown in
Figure 1 (water and freeze-on ice are not considered here). Layers and bedrock are two physical components of the ice sheet subsurface. The ice column consists of a sequence of ice layers. Bedrock is the deepest reflection area of radar wave, and completely attenuates the radar wave. Therefore, the radar equipment under the bedrock can only receive noise. Especially EFZ, which is not a physical component, is the result of the lack of coherent reflector due to the layer disturbance caused by the ice flow at the base interface [
13]. In the EFZ, the reflected wave is buried in the thermal noise, so it has a similar distribution with the noise [
10]. For this reason, in the automatic classification of ice subsurface targets, the EFZ and noise classes are merged within a single no backscattering target class [
10]. Therefore, free space, layers, bedrock, and noise (including EFZ region) are regarded as the classification targets in this paper.
The traditional classification methods need to design features manually, which are high complexity and slow speed, and are not suitable for large, complex datasets. In a more recent development, deep learning technology is employed to better estimate the ice and bedrock boundaries in glacier echograms. Kamangir et al. [
14] combines holistically-nested edge detection (HED) [
15] with the undecimated wavelet transform technique to develop an end-to-end ice boundary detection network. Xu et al. [
16] proposes a multi-task spatiotemporal neural network that combines 3D ConvNets and recurrent neural network (RNN) to estimate ice surface boundaries from sequences of tomographic radar images. However, the identification of the ice subsurface targets using deep learning technology has not been fully explored. Deep learning algorithms are efficient in many public data sets (e.g., Cityscapes dateset, PASCAL-VOC dataset) because they can automatically learn the features of the different data scales without manual intervention. Chen et al. [
17,
18,
19] propose a series of DeepLab networks for image pixel classification, where the latest network, DeepLabv3+, provides the accurate and high resolution results in PASCAL-VOC2012. Yuan et al. [
20] introduces an object context pooling (OCP) scheme and focus on the context aggregation strategy for robust classification. Fu et al. [
21] propose a dual attention network (DANet) to adaptively integrate local features with their global dependencies. Therefore, it is worth exploring strategies for applying these deep learning methods to the classification of the ice subsurface targets.
In this paper, we propose a deep convolution classification network to achieve pixel-level classification of ice subsurface targets. This network is composed of filter processing, encoder and decoder. Our network has been validated on the radargram data set provided by the Center of Remote Sensing of Ice Sheets (CReSIS) from 2009 to 2011, and the results show that our network can outperform an automatic classification system for the ice sheet subsurface targets based on support vector machine [
10], Deeplab networks [
17,
18,
19], object context network [
20], and dual attention network for scene segmentation [
21].
The main contributions of this paper are listed as follows: (1) for the first time in the literature, we introduce the deep convolution network to realize the classification of the ice subsurface targets, and the network realizes the end-to-end processing, which avoids the complex feature engineering for the ice subsurface targets; (2) in the encoder, the modified ASPP structure is used to obtain multi-scale features and improve classification accuracy by removing image level features and changing feature dimensions; (3) in the decoder, a reasonable method of feature fusion is used to solve the problem of long network training and testing times and further improve the accuracy; (4) we use a bilateral filtering algorithm to reduce the speckle noise of radar image to provide the high level of information from radar images.