Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
6 results
Search Results
Item Indic Visual Question Answering(Institute of Electrical and Electronics Engineers Inc., 2022) Chandrasekar, A.; Shimpi, A.V.; Naik, D.Visual Question Answering (VQA) is a problem at the intersection of Computer Vision (CV) and Natural Language Processing (NLP) which involves using natural language to respond to questions based on the context of images. The majority of existing methods focus on monolingual models, particularly those that only support English. This paper proposes a novel dataset alongside monolingual and multilingual models using the baseline and attention-based architectures with support for three Indic languages: Hindi, Kannada, and Tamil. We compare the performance of traditional (CNN + LSTM) approaches with current attention-based methods using the VQA v2 dataset. The proposed work achieves 51.618% accuracy for Hindi, 57.177% for Kannada, and 56.061% for the Tamil model. © 2022 IEEE.Item Support Vector Regression based Forecasting of Solar Irradiance(Institute of Electrical and Electronics Engineers Inc., 2022) Shimpi, A.V.; Chandrasekar, A.; Keshava, A.; Vinatha Urundady, U.PV power is being increasingly popular in terms of distributed energy source and derives its energy from irradiation of the sun. This irradiation differs demographically and needs to be accurately modelled for optimizing the dispatch of the source. Many methods are already in use to forecast the sun irradiation primarily based on Neural Networks and Machine learning techniques. In this paper, Support Vector based prediction is implemented and verified on a set of data. Support Vector Regressor (SVR) is a method of shifting the data points to a hyperplane and finding the correlation between the data samples. Different Kernel functions are used to define the hyperplane and their performance compared. Various combinations of input data is used to obtain the output from the regressor. Prediction metrics are used to determine the efficacy of the algorithm and based on the metrics the worst and best models for forecasting are presented. © 2022 IEEE.Item UnDIVE: Generalized Underwater Video Enhancement Using Generative Priors(Institute of Electrical and Electronics Engineers Inc., 2025) Srinath, S.; Chandrasekar, A.; Jamadagni, H.; Soundararajan, R.; Prathosh, A.P.With the rise of marine exploration, underwater imaging has gained significant attention as a research topic. Under-water video enhancement has become crucial for real-time computer vision tasks in marine exploration. However, most existing methods focus on enhancing individual frames and neglect video temporal dynamics, leading to visually poor enhancements. Furthermore, the lack of ground-truth references limits the use of abundant available underwater video data in many applications. To address these issues, we propose a two-stage framework for enhancing underwater videos. The first stage uses a denoising diffusion probabilistic model to learn a generative prior from unlabeled data, capturing robust and descriptive feature representations. In the second stage, this prior is incorporated into a physics-based image formulation for spatial enhancement, while also enforcing temporal consistency between video frames. Our method enables real-time and computationally-efficient processing of high-resolution underwater videos at lower resolutions, and offers efficient enhancement in the presence of diverse water-types. Extensive experiments on four datasets show that our approach generalizes well and outperforms existing enhancement methods. Our code is available at github. com/suhas-srinath/undive. © 2025 IEEE.Item Detection of arrhythmia from electrocardiogram signals using a novel gaussian assisted signal smoothing and pattern recognition(Elsevier Ltd, 2022) Chandrasekar, A.; Shekar, D.D.; Hiremath, A.C.; Chemmangat, K.The electrocardiogram is a widely used measurement for individual heart conditions, and much effort has been put into automatic arrhythmia diagnosis using machine learning. However, the classification performance is hampered by the use of less representative data in conjunction with traditional machine learning models. This paper proposes a novel algorithm for pre-processing raw Electrocardiogram signals via Gaussian Assisted Signal Smoothing. In this method, the ECG signal is modeled as a low pass component and a weighted sum of Gaussians. The Gaussians are used to model the peak characteristics of the signal, effectively preserving its structure and morphology while eliminating the noise, which is evident by the enhanced peak signal-to-noise ratio of the GASS signal. The R peaks obtained from the Pan Tompkins algorithm are used to extract the heartbeats from the filtered signal using a windowing technique. A cascaded combination of a Convolutional Neural Network and a Quadratic Support Vector Machine is then used to classify the heartbeats. The CNN model has 131,661 parameters, making it much lighter than previously reported works. The MIT-BIH Arrhythmia Database was used for our experiments. Across eleven classes, our results reveal that the model has an accuracy of 97.63% and an average F1 score of 0.9263. In contrast, previous works have primarily focused on a one vs. all or a five-class classification. From a signal processing standpoint, the proposed method offers a promising solution for Signal Filtering and Arrhythmia Classification. © 2021 Elsevier LtdItem InFLuCs: Irradiance Forecasting Through Reinforcement Learning Tuned Cascaded Regressors(IEEE Computer Society, 2024) Chandrasekar, A.; Ajeya, K.; Vinatha Urundady, U.Accurate prediction of solar irradiance is essential for optimizing renewable energy sources in distributed generation systems due to its significant impact on solar power generation. Despite notable advancements, the inherent variability of irradiance presents challenges for existing models. In this article, we introduce a novel approach for irradiance forecasting using a cascaded combination of regressors applied to transformed process variables. Our method utilizes a gradient-boosted decision tree as the primary regressor to generate initial predictions, which are subsequently refined by a support vector regressor acting as an error correction module. Notably, the secondary regressor's kernel, alongside other hyperparameters, is dynamically learned through reinforcement learning with an RNN-based controller. Evaluation results demonstrate that our prediction-correction framework achieves superior performance compared to state-of-the-art approaches, as indicated by RMSE, MAE, and text{R}^{2} score metrics. Thorough comparative analysis highlights the model's enhanced accuracy and its potential for precise irradiance forecasting. © 2005-2012 IEEE.Item A Low-Complexity Solution for Optimizing Binary Intelligent Reflecting Surfaces towards Wireless Communication(Multidisciplinary Digital Publishing Institute (MDPI), 2024) Janawade, S.A.; Krishnan, P.; Kandasamy, K.; Holla, S.S.; Rao, K.; Chandrasekar, A.Intelligent Reflecting Surfaces (IRSs) enable us to have a reconfigurable reflecting surface that can efficiently deflect the transmitted signal toward the receiver. The initial step in the IRS usually involves estimating the channel between a fixed transmitter and a stationary receiver. After estimating the channel, the problem of finding the most optimal IRS configuration is non-convex, and involves a huge search in the solution space. In this work, we propose a novel and customized technique which efficiently estimates the channel and configures the IRS with fixed transmit power, restricting the IRS coefficients to (Formula presented.). The results from our approach are numerically compared with existing optimization techniques.The key features of the linear system model under consideration include a Reconfigurable Intelligent Surface (RIS) setup consisting of 4096 RIS elements arranged in a 64 × 64 element array; the distance from RIS to the access point measures 107 m. NLOS users are located around 40 m away from the RIS element and 100 m from the access point. The estimated variance of noise (Formula presented.) is 3.1614 (Formula presented.). The proposed algorithm provides an overall data rate of 126.89 (MBits/s) for Line of Sight and 66.093 (MBits/s) for Non Line of Sight (NLOS) wireless communication. © 2024 by the authors.