Context: SAR images can be acquired all time (day and night and whatever the weather) which is a real advantage for tropical areas. Nevertheless the signal is affected by speckle noise which induces strong fluctuations. Edge detection in such situations is difficult and usually lead to variable false alarm rate.
Objective: This project is dedicated to edge detection on SAR images. The idea is to train a network on natural images with added speckle noise (for which a ground truth is available [1]), with or without transfer learning and evaluate its performance on simulated and real SAR images. Comparisons with usual gradient for SAR data [2] will also be led (code provided).
Supervision: Florence Tupin, Emanuele Dalsasso
[1] Xie, Saining and Tu, Zhuowen, Holistically-Nested Edge Detection,
Proceedings of IEEE International Conference on Computer Vision, 2015
https://github.com/moabitcoin/holy-edge
[2] F. Dellinger, J. Delon, Y. Gousseau, J. Michel and F. Tupin, SAR-SIFT: a SIFT-like algorithm for SAR images, IEEE Transactions on Geoscience and Remote Sensing , January 2015
