Object Detection

  • Developed a TFLite model for the detection of police cars, bikes, and vans.
  • Deployed the model in android app

Lung Cancer Detection

  • Data Pre-processing: Resized, adjusted the luminance, rescaled, and converted data to uint8
  • Training Nodule Detector: A 3D Faster R-CNN Network is used as the detector
  • Proposals: The model trained was tested on all data, giving all suspicious nodule locations and confidences
  • Training Nodule Classifier: Sampled proposals according to confidence, and for each proposal, cropped cubes, centered at the proposed center

Diabetic Retinopathy Classification

  • Developed a deep learning model to classify Diabetic Retinopathy using CNN and transfer learning
  • Tech-Stack: Python, Keras

Information-Theoretic Feature Selection from Hyperspectral Images

  • Proposed greedy feature selection algorithm using mutual information, Relevancy, Redundancy, and Representability
  • Carried out experiments on two hyperspectral images i.e. Indian Pines and Salinas Valley
  • A considerable increase in accuracy(5-6 %) as compared to baseline accuracy was observed
  • It outperforms existing methods such as mRMR, PCA, and NMIFS even at lower dimensions
  • Tech-Stack: Python, Keras
  • Publication