Jerome Michael Issak

Topic : Semantic Segmentation and Quality Improvement of Synthetic Aperture Radar(SAR) imagery using Deep Learning - Developed 8 Deep Neural Network (DNN) models for segmentation of surface classes in radar satellite images. - Despeckled SAR images for noise reduction using pre-trained Convolutional Neural Networks. - Achieved an accuracy of 97% on test images and deployed the best performed model in a docker container.

Topic : Road Surface Pothole Detection using Deep Neural Networks - Designed a deep learning model for detecting road damages using camera images for improving road conditions. - Potholes are detected with 95% accuracy on test images using transfer learning approach. - Optimized the best model using Tensor RT library and deployed in Nvidia Jetson Tx2 with a run-time inference speed of 25 FPS.