Speaker
Dr
Armando Marino
(The Open University)
Description
SAR is very valuable for monitoring icebergs due to its capability to acquire images with almost any weather condition and without solar illumination. In SAR images, icebergs are often visible as bright regions or points (depending on iceberg size and image resolution). This work is focused on the detection of smaller icebergs or thick/deformed blocks of sea ice embedded in thinner sea ice. The detection of icebergs of several km size is routinely done, however, there are still issues in identifying bergy smaller than a few hundred meters, especially when embedded in sea ice. In fact, in some situations sea ice can appear very bright defeating detectors based on image intensity. The use of polarization is expected to improve the detection capability of the system. The aim of this work is to propose a novel detector based on dual-polarimetric detected images (i.e. only intensity) to improve the detection of small icebergs embedded in sea ice.
The proposed algorithm is based on dual-polarimetric detected images (i.e. only the intensity of the images is necessary, not the phase). The physical idea behind the algorithm is that icebergs or thick/deformed ice are supposed to have a stronger volume contribution compared to thin sea ice. Moreover, we are interested in detecting icebergs with a well-defined size. Two boxcar filters are applied over the entire HV and HH images, exploiting two different window sizes (a test and training window). The intensities obtained by these averaging are arranged in a compact mathematical expression:
$\Lambda = \frac{|\langle | HV |^2 \rangle_{test}-\langle | HV |^2 \rangle_{train}|}{\langle | HH |^2 \rangle_{train}}>T$
As mentioned previously, there is a specific size of the targets (icebergs or thick/deformed ice) that are able to trigger the detection. In order to change the dimension of the target of interest, the windows have to be modified. Clearly, we cannot be completely sure if the detected object is an iceberg or a block of thick/deformed ice, but they both may represent hazards for the navigation. Additionally, the proposed detector is more focused on relatively small bergs and it may be ineffective for large icebergs of several km. In such cases a more traditional methodology based on morphological characteristics of the target is suggested.
In order to test the detector, real Sentinel-1 dual-polarimetric images are exploited. The data were acquired in Greenland, near the basin of the Helheim glacier. The latter is one of the fastest calving glaciers. Moreover, the acquisitions were performed in cold seasons, where the sea is covered in sea ice.
Unfortunately, ground survey of the icebergs or thick/deformed ice is not available and we have to rely on visual inspection of the images. In particular, targets of interest were identified as bright regions in the HV channel that cast a shadow in the far range and a bright rim in the near range.
The proposed algorithm is able to detect icebergs of dimensions chosen by the selection of the averaging windows. Finally a comparison with algorithms using HV intensity alone shows that there is a much reduced number of false alarms.
Primary author
Dr
Armando Marino
(The Open University)
