Work

Projects & Experiences

  1. Thesis: Knowledge-based semantic enrichment for semantic image segmentation (ongoing)
    Rajeswari Parasa
    My master's thesis project evaluated the impact of incorporating knowledge in deep learning models for image segmentation tasks in Remote Sensing using a semantic enrichment of Sentinel-2 multispectral images. Particularly, it uses a pixel-wise land cover classification task to analyse the impact. The project involves developing PyTorch workflows that use semantically enriched image patches, train a deep-learning model to perform a landcover classification task and obtain inferences on unseen patches for model evaluation. The project conducts experiments in both training-from-scratch and pretraining settings. In the trained-from-scratch models, the results show a notable increase in the performance of the deep model with knowledge-based semantic enrichment. However, in the pretraining settings, the results were inconclusive of the usefulness of semantic enrichment, potentially due to the limited size of the dataset used for pretraining. However, this can be easily overcome by using a much larger number of Sentinel-2 patches, automatically obtaining the corresponding semantic enrichment from the SIAM tool, and repeating the tests shown in this project for further evaluation of pretraining settings. The study showcases the potential value of SIAM-based enrichment for deep learning models in Remote Sensing applications.
    Code Repo, Full Report

  2. Timeseries Classification for Crop Identification using LSTM
    Rajeswari Parasa
    This project demonstrates time-series classification for crop identification on a subset of the MiniTimeMatch dataset by training a deep learning model with an LSTM architecture. The notebook showcases data exploration, setting up a data preparation pipeline using a custom dataset class, and finetuning the model architecture and the training loop. Code showing usage of trained models for inference on new data in also included.
    Code Repo

  3. Housing Price Prediction
    Rajeswari Parasa & Candela Sol Pelliza
    In this exercise, we build and test various regression models to predict housing prices in Ames, Iowa. The project produced a set of notebooks detailing data exploration and preparation steps, followed by a demonstration of three regression models, with different configurations, applied to the preprocessed data. These different models include Basic Linear Regression, Linear Regression with ElasticNet Regularization, Random Forest Model, Random Forest with Boruta Feature Selection, and Catboost Model. The notebooks also demonstrate workflows for hyperparameter tuning for each of the models.
    Code Repo

  4. Summer'23 Internship
    Rajeswari Parasa
    In the summer of 2023, I interned with the EO Analytics team of the ZGIS department at the University of Salzburg, Austria. Under the guidance of Assoz. Prof Dirk Tiede and Dr Martin Sudmanns, I developed a series of automated workflows to query EO data cubes generated using sen2cube.at infrastructure built on Sentinel-2 data and its semantically enriched data. The workflows are aimed at exploring the potential of EO data cubes in building ESG indicators for monitoring real-estate properties. More specifically, I propose a 'green score' that can be monitored over time to assess the green cover and its changes on a property parcel. I also build a workflow for retrieving the 'semantic history' of a property parcel. These two exercises hint at the potential of EO data cubes in monitoring and assessing the sustainability of real-estate properties.
    Summary Report, Paper

  5. Interactive tool for visualising Delhi Draft Master Plan 2041
    Nikhil VJ and Rajeswari Parasa
    We built the tool to make the spatial data of the draft plan more accessible to researchers and citizens. The tool was used by the Main Bhi Dilli campaign as part of their work to raise awareness about the plan among people.
    Tool, Code Repo

  6. Exploring spatial routing libraries in python
    Rajeswari Parasa
    In this notebook, I explore Networkx & OSMNx, OSRM (free and open source) and Google Maps Direction API for finding street network based routes between a set of origin-destination (O-D) points and compare them visually.
    About, Code


  7. ## Working Papers and Reports
  8. Methods to measure spatial access to healthcare facilities in cities: A case study of the urban poor in Chennai (2021)
    Rajeswari Parasa, Harsh Vardhan Pachisia and Isalyne Gennaro
    The study explores open-source tools and libraries to analyse geographic distances from slums to the nearest primary healthcare centres in an Indian city. It argues for undertaking similar geospatial analysis while planning new health facilities.
    Download Paper

  9. An investigation of National Open Government Data Platforms: How can India improve? (2021)
    Sridhar Ganapathy, Harsh Vardhan Pachisia, Rajeswari Parasa and Isalyne Gennaro
    The paper studies and evaluates national open government data (OGD) platforms of several countries and the policy frameworks that guide the creation and use of these platforms. It makes recommendations for India's OGD platform.
    Download Paper

  10. The State of Open Data in Urban India (2022)
    Harsh Vardhan Pachisia, Anushka Bhansali and Rajeswari Parasa
    The paper investigates the urban open data landscape in India by looking at the stakeholders and aspects of data quality and accessibility.
    Link

  11. Unlocking Potential of India’s Open Data (2022)
    A NASSCOM initiative in collaboration with Fractal, Microsoft, Infosys, IDFC Institute, TCS & Amazon
    Link

  12. Contribution of infection and vaccination to seroprevalence through two COVID waves in Tamil Nadu, India (2021)
    T.S. Selvavinayagam, A. Somasundaram, Jerard Maria Selvam, Sabareesh Ramachandran, P. Sampath, V. Vijayalakshmi, C. Ajith Brabhu Kumar, Sudharshini Subramaniam, K. Parthipan, S. Raju, R. Avudaiselvi, V. Prakash, N. Yogananth, Gurunathan Subramanian, A. Roshini, D.N. Dhiliban, Sofia Imad, Vaidehi Tandel, Rajeswari Parasa, Stuti Sachdeva, Anup Malani
    Link

  13. SARS CoV-2 seroprevalence in Tamil Nadu in October-November 2020 (2021)
    Anup Malani, Sabareesh Ramachandran, Vaidehi Tandel, Rajeswari Parasa, Sofia Imad, S. Sudharshini, V. Prakash, Y. Yogananth, S. Raju, T.S. Selvavinayagam
    Link


  14. ## Short Pieces
  15. Embed open data principles in Master Plans to make planning more inclusive and participatory
    Rajeswari Parasa and Harsh Vardhan Pachisia
    In the context of Delhi Master Plan 2041, we argue for opening up spatial datasets generated through city plans to improve public participation.
    Link

  16. How Well Did Bed Capacity Respond To Patient Needs In Mumbai?
    Nikita Kwatra and Rajeswari Parasa
    In this piece, we analyse COVID-19 bed capacity and occupany data published by Municipal Corporation of Greater Mumbai to better understand the system's response to the surge of infections over time. We also look at the usefulness of monitoring test positivity rate to better prepare for upcoming infection surges and demand for hospital beds.
    Link

  17. Status Quo of Spatial Datasets of Health Facilities in India
    Rajeswari Parasa
    The piece looks at existing openly available spatial datasets of health facilities in India, highlights gaps and the immediate need to fill those gaps.
    Link