Research projects completed
1. Improving the performance of optimizers in deep learning May 2022 – May 2023
Funder: School of Computer Science and Engineering, Kyungpook National University, South Korea.
Objective(s):
- Improve model reliability with deep neural networks by accelerating the convergence of gradient descent optimizers.
- Improving non-linearity representation to model complex relationships by modifying the activation functions in deep learning.
Role: Project Lead
Responsibilities:
- Designing deep learning models for evaluating the proposed idea.
- Incorporating the proposed ideas into TensorFlow by overriding the architectural modules.
- Develop mathematical proofs and theoretical analyses of the intended idea.
- Setting up a distributed cluster and a sharable GPU for conducting the experiments.
- Conducting experiments and performing empirical analyses and ablation studies.
2. Abnormal image detection and visualization of fabric products Sep 2022 – May 2023
Funder: National Research Foundation (NRF)
Objective(s):
- Clustering the anomaly objects and localizing the anomaly present in the industrial fabrics.
Role: Project Associate
Responsibilities:
- Performed data augmentation techniques to increase the number of training images for the model.
- Implemented pre-processing methods such as noise removal and contrast enhancement to improve image quality.
- Contributed to model development by exploring autoencoders for anomaly localization.
- Evaluated model performance on an anomaly detection dataset and supported analysis.
- Documented experimental details and results in project reports.
3. Real-time fault detection in fused deposition modelling July 2022
Funder: Department of Mechanical Engineering, National Institute of Technology Calicut, India in collaboration with School of Computer Science and Engineering, Kyungpook National University, South Korea.
Objective(s):
- Real-time error detection in fused deposition modelling using CNN.
Role: Project Associate
Responsibilities:
- Collecting sample images under different error categories (staircases, overfill, and void defects) during fused deposition modelling.
- Generating additional images using data augmentation techniques to avoid overfitting.
- Devising a suitable CNN model to train and classify defects.
- Comparing the proposed method with standardized classification algorithms.
4. Gesture recognition system for sign language and music generation Feb 2022 – May 2023
Funder: School of Computer Science and Engineering, Kyungpook National University, South Korea.
Objective(s):
- Real-time music generation based on hand gesture recognition.
- Sign language detection through gestures.
Role: Project Coordinator
Responsibilities:
- Modified GRU model to enhance gesture recognition accuracy and performance.
- Developed a validation framework to ensure model effectiveness across various gesture types for maintaining project quality standards.
- Coordinated the technical integration of deep learning models with application interfaces, streamlining the real-time translation of recognized gestures into output.
- Managed technical documentation and project guidelines, supporting project scalability, and ensuring knowledge transfer for future development.
- Facilitated collaboration between technical and user experience teams, bridging the gap to optimize user interface design for gesture-based applications, enhancing overall project cohesion and execution.
5. Korean handwritten text recognition Dec 2021 – Feb 2022
Funder: Small and medium business technology information promotion agency, Daegu, South Korea.
Objective(s):
- Generating Korean text from old-fashioned Korean handwriting.
Role: Project Technical Support Specialist
Responsibilities:
- Assisted in annotating bounding boxes around handwritten text.
- Engaged in image pre-processing, applying algorithms based on the histogram of gradients to extract features from the handwritten text.
- Supported the development of the Bi-LSTM model for generating machine-encoded text.
- Helped evaluate the model’s performance, utilizing metrics like character error rate to rigorously assess OCR accuracy.
6. Small object detection in manufacturing and smart city Mar 2021 – Feb 2022
Funder: Small and medium business technology information promotion agency, Daegu, South Korea.
Objective(s):
- Identification of illegal parking in the disabled car parking slot.
- Identification of improperly placed small objects (stickers/badges).
Role: Project Associate
Responsibilities:
- Conducted on-site data collection by visiting car parking places to capture photos of disabled car parking slots for model training.
- Pre-processed and annotated the collected images to identify illegal parking and small objects (stickers/badges) inaccuracies.
- Assisted in finding an appropriate deep learning algorithm for automated detection of illegal parking and improperly placed objects.
- Performed iterative testing and model adjustments based on initial results and feedback to enhance detection accuracy and system efficiency.