Software

My research interest is in SAR, PolSAR and InSAR image and signal processing for environmental monitoring. Therefore, I have developed several methods, workflows and architectures to perform extract, load, transform tasks in this domain. Here are some selected developments in the mentioned areas.

Sentinel-1 SAR backscatter analysis ready data preparation in Google Earth Engine

I am the main developer and maintainer of the gee_s1_ard framework developed in GEE Javascript and Python API used to process Sentinel-1 SAR ground range detected images to the analysis ready data level. The scripts perform additional border noise correction, mono-temporal and multi-temporal speckle filters and radiometric terrain flattening.

SAR image despeckling

I have developed deep learning based methods to despeckle single channel SAR image despeckling deSpeckNet in the real number domain

and PolSAR image despeckling in the complex domain CV-deSpeckNet.

SAR and optical time-series classification

I have also developed methods to fuse SAR and optical time-series data for the detection of forest disturbance. I used recurrent neural networks such as LSTM to implicitly learn the seasonality pattern of time-series signals in tropical dry forests.

SAR images semantic segmentation

I have also developed methods to perform semantic segmentation for SAR image time-series data SAR-FCN-DK3. I used a fully convolutional network with dilated kernels for landcover classification.