Omm Prakash

Experience

• 

  Indian Institute of Technology (Banaras Hindu University), Varanasi

  Dec 2021 - Apr 2022

  Automatic Faults Detection of Photovoltaic Farms using Thermal Images

  With the increasing demand for renewable energy, proper management of photovoltaic hubs is essential. But fault detection in them is still a challenge. The central temperature consequenced faults can be detected using Thermal analysis of PV modules. The Robotics and Artificial Intelligence Department of Information Engineering at Università Politecnica Delle Marche in Italy provided the photovoltaic thermal image dataset by UAV copter. It was used in conjunction with a properly hyper-tuned model.For detections, we used YOLOv5 (YOU LOOK ONLY ONCE version 5), a novel state-of-the-art CNN that detects objects in real-time with high accuracy. This method employs a single neural network to process the entire image, divide it into parts, and predict each component's bounding boxes and probabilities. The proposed model(in the future) would be deployed in jetson which will be used as a companion computer in a UAV Copter, and it would be used on our college campus. Show less

• 

  DoDAO

  Jul 2022 - Dec 2022

  Blockchain Intern

  I gained a better understanding of how the blockchain-based industry works and how effective it is through my internship at DoDao-io. I interned at DoDao and explored various Web3-based technologies like DiFI, Aave, etc. I had a great time working with the Solidity Content development team to create and publish multiple blockchain insight content. I learned much about blockchain technology, development, and related applications. However, I still have a long way to go in understanding the more technical aspects of web3 and the application of Machine learning in it. I enjoyed my internship experience and am confident I can apply the skills I learned in my future career. Show less

• 

  Indian Institute of Technology (Banaras Hindu University), Varanasi

  Sept 2022 - now

  I preprocessed hyperspectral images, with dimension reduction to clean the newly created Microscopic Hyperspectral Choledoch Dataset of size 102 GB. The database contains microscopy hyperspectral and color images of cholangiocarcinoma, including 880 scenes from 174 people. Nvidia Tesla T4 GPU is used for fast computation.Image preprocessing on such a large dataset was the main challenge in the research project. We used 3D CNN to detect the location of cancer cells in an HSI image. The Super Computer is used for fast parallel computing, handling the entire data, and training the model. Show less I am working on the CMU CERT dataset, which is specifically designed for use in cyber security. The dataset must be more balanced, resulting in better model performance on minor classes. I used Pytorch to implement various GAN models on sequential data to improve their balance, such as SeqGAN, RankGAN, CoTA, and others. I am also investigating deep learning architectures for sequential data, such as Multi-Head and Additive Attention, Transformers, LSTM, and so on, and implementing the same after modification for the classification model using TensorFlow. Show less

  • Cancer Cell Detection Using Deep Learning

    Dec 2022 - now

  • Insider Threat Detection

    Sept 2022 - now