Research

Thesis

KUNet — a modified genetic-algorithm-based CNN for Bengali Sign Language detection

KUNet is a convolutional neural network that detects Bengali sign language, with model parameters optimized by a genetic algorithm. Because genetic algorithms are slow learners, we distributed the search across many machines to make training tractable.

Publications & datasets

An Automatic Computer-Based Method for Fast and Accurate Covid-19 Diagnosis

A sequential CNN that detects COVID-19 from CT scan images with ~92.5% accuracy.

DOI: 10.1101/2020.07.02.20136721

KU-HAR: An Open Dataset for Human Activity Recognition

18 activities, 90 participants, smartphone accelerometer + gyroscope. 1,945 raw samples and 20,750 subsamples.

DOI: 10.17632/45f952y38r.5

KU-BdSL: Khulna University Bengali Sign Language dataset

1,500 images covering the 39 consonants of the Bengali alphabet, from 37 volunteers, processed for ML use.

DOI: 10.17632/scpvm2nbkm.1

Education

BSc in Electronics and Communication Engineering — Khulna University, 2017-2023.