Projects
Computer Vision Project at ISRO
Synthetic Aperture Radar (SAR) is the radar technology used by satellites to capture high resolution images of the Earth’s topography. It also has application in body imaging. One of the advantages of using radar over optical sensor is that it can capture images both during day and night-time. Another obstacle overcome by the radar is that it does not get blocked by cloud cover or other atmospheric artifacts such as smoke or smog that can affect the visibility of objects or land features in the satellite feed. However, one disadvantage of using ISRO’s RISAT satellite is that a high exponential noise is produced in the images which makes it extremely grainy leading to subsequent information loss. Thus, in this project I have proposed a state-of-the-art despeckling module which uses wavelet transform and Principal on Approximation Coefficients (POAC) algorithm to remove noise from SAR Images.
Deep Learning Systems Project
This project attempts to use satellite images and
other remotely sensed imagery to perform large-scale image segmentation. The project implements a slightly tweaked version of a U-net model which is classically used in medical imaging based applications for this task as past literature indicates the possibility of being able to successfully perform this task in the event that the availability of data is a choke-point. Further it aims to perform a natural event prediction task (using a binary classifier that predicts whether an event has taken place or not) by exploiting the pretrained segmentation model.
Computer Vision Project
The goal of the project is to have a state-of-the-art model for both classification and segmentation of organ cells to accurately determine the organ based on the electron microscope sliced image. For the image segmentation task, the model used is the Unet architecture as it is popularly used in biomedical image segmentation and is reviewed to give impeccable performance on biomedical datasets. For the classification task, the popular ResNet architecture and its variants are used to draw comparative analysis among the models.
Computer Vision Project
This project aims to solve a time series prediction problem through an LSTM deep learning model built in PyTorch. The project aims at predicting the third wave of Covid-19 pandemic in India based on historic data collected from WHO website. The goal of the project is to help the healthcare domain to be better prepared for an upcoming pandemic wave, through logistic arrangement of medicines, vaccines, hospital beds etc. Getting the prediction for the number of cases can help prevent disasters of mismanagement.
Time Series Project at IU University
The Time Series Analysis component focuses on understanding homeownership trends across various states in the United States. The primary goals were to train a sophisticated time series prediction model and detect outlier states displaying significant deviations from the national homeownership rate.
