- Topic: Iceberg and ship detection in satellite imagery
- Goals: The project goal is to build an algorithm for the detection of ships and icebergs in Sentinel-1 SAR imagery. Desired output is a map, which shows the locations of icebergs, ships and unidentified objects.
- Details: The dataset used for training is obtained from a Kaggle challenge, Statoil/C-CORE Iceberg Classifier. Each image has 75x75 pixels with two bands from HH and HV polarisations and contains a ship or an iceberg. This dataset will be used to train a CNN. After training the classification model, we will use Sentinel-1 SAR images to show the "real world application" of our model. The satellite images will be pre-processed with the Sentinel Application Platform (SNAP) Python API. We will then identify bright objects within each satellite image. A 75x75 subset of the radar image will be made for each object and fed into our classification model. Finally, the results will be plotted on a map.
- Disko Bay
-
S1A_IW_GRDH_1SDH_20210115T100027_20210115T100052_036147_043CF4_049C
- Other Type/Auxillary N69°08.750' W53°39.933'
- Fishing Vessel N68°45.080' W51°20.846'
-
S1B_IW_GRDH_1SDH_20210114T100803_20210114T100828_025149_02FE89_234D
- Other Type/Auxillary N69°05.990' W53°18.654'
- Fishing Vessel N68°51.626' W52°47.673'
- Other Type/Auxillary N69°23.972' W51°36.317'
- Fishing Vessel N68°43.877' W51°30.219'
- Fishing Vessel N68°43.825' W51°21.114'
- Cargo Ship N76°28.180' W54°08.052'
-
- Svalbard
- S1B_IW_GRDH_1SDH_20210108T154500_20210108T154525_025065_02FBC6_38D2 (contains only a tanker and no icebergs)
- Tanker N78°12.417' E14°32.650'
- S1B_IW_GRDH_1SDH_20210108T154500_20210108T154525_025065_02FBC6_38D2 (contains only a tanker and no icebergs)
-
Short article and code of another solution for the kaggle contest on towardsdatascience.com
-
Ship-Iceberg Classification
- Using SAR and Multispectral Satellite Images with SVM and CNN (Heiselberg 2020)
- Using Sentinel-1 SAR images for Object Detection and Classification (Heiselberg 2020)
Notes
Calculating total backscatter and cross-polarisation ratio:
total backscatter: H = HH HV
cross-polarisation ratio: C = HV / H
- Presentation of a detailed Jupyter Notebook with code and comment
- including the definition of the environment
- including required sections (Introduction, Data and Methods, Results, Baseline)
- A small video, accompanying, for example, a screen recording of the notebook with an explanation of the challenge of the project, the used approach, and the results.
- A statement that the code is released as open source software. The data you use in your project can remain private if you wish.
- Time: 8 -- 10 minutes