Dr. Kumaresh Sheelavant
@saividya.ac.in
Assistant Professor,
Scopus Publications
- Active Learning Pipelines for Efficient Annotation of Large-Scale IOT Video Streams
Kumaresh Sheelavant, Sanjay Kumar Suman, G Sravanthi, L. Bhagyalakshmi, J Jesintha Princy, D. Sravanthi
Proceedings of the 4th IEEE International Conference on Interdisciplinary Approaches in Technology and Management for Social Innovation Iatmsi 2026, 2026
Manual frame-level annotation for large-scale IoT video streams in Indian urban deployments imposes prohibitive cost and latency, undermining timely analytics for traffic management and incident response. This work presents an edgeaware active-learning pipeline that fuses epistemic uncertainty sampling with geometric/topological diversity through persistence-based summaries of short embedding windows and temporal-consistency regularizers. The pipeline is engineered for low-latency on-device operation and amortized topological computation. Evaluation uses an India urban traffic video corpus and a ResNet-18 / video-embedding backbone as a pragmatic baseline (ResNet-18 edge pipeline and dataset split reported in the reference). Empirically, the hybrid acquisition reduced annotation expenditure by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\approx 40-60 \%$</tex> to reach operational F1/AUC comparable to full supervision, and with 50 % of labels achieved AUC up to 0.998 (marginally exceeding the baseline AUC <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{\approx} \mathbf{0. 9 9 7}$</tex>) while preserving sub-second decision windows through model compression and batch amortization. The results indicate that topologically-aware, temporally-aware acquisition materially increases per-label utility and enables practical, privacy-preserving municipal rollouts that minimize human-in-the-loop burden. - Real-Time Offline Edge AI Framework for Sensor Integrated Precision Agriculture
T G Manjunath, Sukruth Gowda M A, Kumaresh Sheelavant, Harika Harika, Itikala Sriram, Monika M
Proceedings of the 2026 International Conference on AI Driven Smart Systems and Ubiquitous Computing Icauc 2026, 2026
Real-time agricultural decision-making in small-scale and rural farming environments remains challenging due to unreliable connectivity, fragmented toolchains, and the absence of localized advisory systems. This paper presents a real-time offline edge AI framework for sensor-integrated precision agriculture, in which crop classification and fertilizer recommendation are performed entirely on-device without reliance on cloud infrastructure. The proposed system integrates an Arduino-based soil sensing unit capable of measuring Nitrogen (N), Phosphorus (P), Potassium (K), soil pH, temperature, humidity, and moisture with a lightweight edge-based machine learning pipeline. A supervised multi-class crop prediction model trained and evaluated on a curated Indian agro-climatic dataset consisting of 52,001 records across 13 crop classes achieves an accuracy of 96.28%, with macro-averaged precision, recall, and F1-score of 0.96. Unlike prior approaches that address crop estimate or fertilizer recommendation in isolation, the proposed framework unifies crop classification, dataset-driven fertilizer advisory, and rule-based disease risk alerts within a single fully offline system. Predictions are logged locally in a lightweight JSON store and visualized using pre-downloaded geographic boundaries, enabling effective real-time recommendations in low-resource and connectivity-constrained deployment settings. - Improving Transparency and Adaptability in AI with Hybrid Generative Adversary Attention Networks
Kumaresh Sheelavant, Lakshmi Chandrakanth Kasireddy, Nelli Sreevidya, V. Arunkumar, L. Jayanthi, Hitha Poddar
Research Advance S in Intelligent Computing Volume 3, 2026
Artificial intelligence (AI) has improved tremendously to support intelligent decision processes which now exist in healthcare and finance as well as autonomous systems. Current AI systems have three key limitations, which include restricted performance under reduced datasets and the inability to explain their functions, alongside their failure to adapt well to changing operational conditions. The proposed solution to these problems uses Hybrid Attention-Based Generative Adversarial Adaptive Network for Explainable AI (HAGAN-X), which represents a new algorithm. The HAGAN-X system generates excellent synthetic data using Generative Adversarial Networks (GANs) and employs methods of attention alongside adaptable learning processes to maximize efficiency in real-time. A built-in explainable AI (XAI) module in HAGAN-X uses interpretable visual and textual outputs to deliver transparent decision explanations to users. The implementation of this combined method allows for more precise and stable artificial intelligence systems, which provide transparent operation when dealing with uncertain conditions. Numerous benchmark assessments across diverse areas indicate that this method surpasses the latest algorithms by exhibiting superior accuracy, enhanced adaptability capabilities, and increased accessibility in analyses of varied datasets. The unified framework of HAGAN-X has proven its ability to extend modern AI capabilities because it achieves equilibrium between intelligence capabilities, adaptive features, and interpretability benefits. The method provides a future-oriented approach to generate AI systems with dual capabilities of power and human-focused accountability. The HAGAN-X model attained an overall accuracy of 97.0%, surpassing all other current methodologies in the assessment. This high accuracy underscores its efficacy in providing precise and dependable estimates across numerous environments. - Deep Neural Network to Enhance Natural Image Structure for an Inverse Problem
Kumaresh Sheelavant, T G Manjunath, Abhinaya Prakash, Dhanya V Tejaswini, Chandana S, Deepthi S
2025 1st International Conference on Advancement in Futuristic Technologies Icaft 2025, 2025
Inverse problems in image processing aim to reconstruct original images from degraded or incomplete measurements such as blurred, noisy, or under sampled data. While traditional approaches rely on handcrafted priors and iterative optimization, Deep Neural Networks (DNNs) have transformed this domain by learning natural image structures directly from data. By implicitly capturing complex statistical priors, DNN-based methods enable high-quality reconstructions that preserve textures, edges, and perceptual fidelity. The solution proposed in this work exhibits significantly high performance over existing techniques, achieving a Peak Signal to Noise Ratio (PSNR) of 58.0 dB, indicating minimal reconstruction error, a Structural Similarity Index (SSIM) of 0.998, demonstrating near-perfect structural similarity, and a pixel accuracy of 99.7%. These parameters clearly illustrates that our approach preserves visual quality far better than previous techniques. The solution proposed sets a new benchmark, while other models show significantly lower performance across all metrics. The proposed method gives a detailed overview of how deep learning promotes natural image structure in inverse problems, discussing key architectures, training strategies, applications, and challenges. - A Neural Network Approach for Speech to Text Translation of Low Resource Language
Kumaresh Sheelavant, Preksha G P, Sakshi Shanbhag, Shreya B C, Sneha T, Shilpa Patil
2025 1st International Conference on Advancement in Futuristic Technologies Icaft 2025, 2025
Developing speech translation systems for low-resource languages remains a major issue because lack of annotated datasets and linguistic resources. This work introduces a neural speech-to-text translation framework designed for low-resource Indian languages, with Konkani as a case study. The proposed system converts spoken Konkani into English text through Hindi as a pivot language, integrating Automatic Speech Recognition (ASR) and Neural Machine Translation (NMT) models within a unified architecture. The methodology involves audio preprocessing, feature extraction, and pivot-based translation using pretrained models such as IndicConformer and Transformer-based NMT to ensure fluency and adaptability. By leveraging transfer learning and multilingual modeling, the approach effectively addresses issues of data scarcity, accent variation, and linguistic diversity. In comparison, our proposed pipeline attains 29.6% WER and 28.1 BLEU, demonstrating competitive performance despite operating in a low-resource setting. These results show that our model performs on par with, and in some cases surpasses, existing benchmarks for similar multilingual pipelines. - DRA-Net with ISSA: A High-Performance Intrusion Detection Framework for IoT Networks with Enhanced Accuracy and Reduced False Alarms
P.Mishra, T.G.Manjunath, A.C.Vikramathithan, K.Sheelavant, N.Nalinakshi, P.K. Pareek
International Conference on Electrical Computer and Energy Technologies Icecet 2025, 2025 - Secure Object Identification Techniques for Autonomous Vehicles
Kumaresh Sheelavant, Sarika S Shirolkar, Anusha Suresh Mysore, S R Indu, Srujala Shetty, Nisha S K
2025 1st International Conference on Advancement in Futuristic Technologies Icaft 2025, 2025
Secure object identification is a cornerstone of safety-critical perception systems in Autonomous Vehicles (AVs), particularly when navigating unstructured environments. However, detection accuracy often degrades significantly under adverse weather conditions such as fog, rain, haze, and low light, posing severe risks to operational safety. This paper investigates these challenges and proposes a detection framework leveraging the YOLOv8 architecture to enhance visibility and classification reliability. The methodology utilizes the Dawn dataset, specifically curated for adverse weather scenarios, focusing on six high-priority classes: cars, buses, trucks, pedestrians, motorcycles, and bicycles. Through rigorous preprocessing, transfer learning, and a training regimen of 130 epochs, the proposed model demonstrates a superior trade-off between inference speed and detection accuracy compared to traditional multi-stage detectors. The results highlight the system’s capability to mitigate false negatives in low-visibility frames, validating YOLOv8 as a viable solution for secure AV navigation. The paper concludes by outlining future research directions, including dataset expansion and multimodal sensor fusion, to further address class imbalances and environmental complexities. - Optimal routing and scheduling for cognitive radio sensor networks using ensemble multi probabilistic optimization and truncated energy flow classification model
Kumaresh Sheelavant, Sumathi R, Charan K V
International Journal of Engineering Trends and Technology, 2021 - Interference and Delay Aware Routing Algorithm for Cognitive Radio Sensor Networks
Kumaresh Sheelavant, R. Sumathi
2017 2nd International Conference on Emerging Computation and Information Technologies Icecit 2017, 2018
Cognitive Radio Sensor Networks (CRSN) consists of cognitive devices capable of changing their transmission parameters on a real time, based on the spectrum available in the environment. These capabilities brings-up the possibility of designing flexible and dynamic spectrum strategies with the purpose of opportunistically accessing the portion of the spectrum temporarily vacated by the primary users, due to this there will be increased complexity in the design of communication protocols. The characteristic of CRSN that is opportunistic access of spectrum raises interference problem in the communication networks. On account of this problem the performance of the network will be degraded like increase in consumption of energy, switching delay, reducing the reliability of a network. To overcome all these challenges we developed an interference and delay aware routing algorithm which selects a best path to route the data from source to the destination. - Energy efficient reliable routing through dynamic spectrum management in cognitive radio sensor networks
Kumaresh Sheelavant, R. Sumathi
Proceedings of 2014 International Conference on Contemporary Computing and Informatics Ic3i 2014, 2014
Cognitive Radio (CR) offers promising solution for spectrum scarcity problem by means of Dynamic Spectrum Management. Routing in Wireless Sensor Networks (WSNs) was a very big issue because of opportunistic access of spectrum band as well as dynamic variable change in environment. Previous research on routing in WSNs are centralized or distributed, but these techniques do not provide high reliability in routing Highly Delay Sensitive Data (HDSD). Instead, it increases the packet delay, interference, spectrum handoffs, as well as increase in energy consumption. In this paper, we developed an Energy Efficient Reliable Routing (EERR) protocol for Cognitive Radio Sensor Networks (CRSN) that routes the HDSD in a highly reliable manner through cognitive radio sensor nodes using licensed bands to meet the delay bound of an application, and also it sets up a quality routes by using the proper channel allocation technique in a dynamic variable environment. This routing protocol balances the energy consumption problem, eliminates conflicts between the nodes, reduces the routing overhead and channel interference, divides the traffics over different channels and time slots.
RECENT SCHOLAR PUBLICATIONS
- Active Learning Pipelines for Efficient Annotation of Large-Scale IOT Video Streams
S Kumaresh, S Sanjay Kumar, S G, B L, JP J, S D
IEEE International Conference on Interdisciplinary Approaches in Technology … , 2026
2026 - Semantic Fact-Checking and Misinformation Analysis Using LLM and Generative AI
M T G, Jayshri, S Kumaresh, H K, SP Durga, M K
5th International Conference on Sentiment Analysis and Deep Learning (ICSADL … , 2026
2026 - A Neural Network Approach for Speech to Text Translation of Low Resource Language
K Sheelavant, P G P, S Shanbhag, S B C, S T, S Patil
1st International Conference on Advancement in Futuristic Technologies … , 2026
2026 - Deep Neural Network to Enhance Natural Image Structure for an Inverse Problem
K Sheelavant, TG Manjunath, A Prakash, DV Tejaswini, C S, D S
1st International Conference on Advancement in Futuristic Technologies … , 2026
2026 - Secure Object Identification Techniques for Autonomous Vehicles
K Sheelavant, SS Shirolkar, AS Mysore, SR Indu, S Shetty, N S K
1st International Conference on Advancement in Futuristic Technologies … , 2026
2026 - Improving Transparency and Adaptability in AI with Hybrid Generative Adversary Attention Networks
S Kumaresh, K Lakshmi Chandrakanth, S Nelli, A V, J L, H Poddar
Research Advances in Intelligent Computing 3, 396-410 , 2026
2026 - Real-Time Offline Edge AI Framework for Sensor Integrated Precision Agriculture
TG Manjunath, S Gowda, K Sheelavant, H Harika, I Sriram
2026 International Conference on AI-Driven Smart Systems and Ubiquitous … , 2026
2026 - Fuzzy Temporal Rule-Based Natural Language Processig for Medical Text Analysis for Caring for Geriatric People
K Sheelavant, etc
International Journal of Basic and Applied Sciences 14 (Sl-1), 480-487 , 2025
2025
Citations: 7 - DRA-Net with ISSA: A High-Performance Intrusion Detection Framework for IoT Networks with Enhanced Accuracy and Reduced False Alarms
P Mishra, TG Manjunath, AC Vikramathithan, K Sheelavant, N Nalinakshi, ...
2025 5th International Conference on Electrical, Computer and Energy … , 2025
2025 - Smart Gardening System with IoT Sensors and AI-Powered Plant Care
K Sheelavant
2025 - Ensemble Learning-Based Intrusion Detection and Classification for Securing IoT Networks: An Optimized Strategy for Threat Detection and Prevention
K Sheelavant
Journal of Intelligent Systems and Internet of Things 17 (2), 101-118 , 2025
2025
Citations: 39 - A SMART WEAPON DETECTION TECHNIQUE USING ARTIFICIAL INTELLIGENCE AND DEEP LEARNING FOR SECURITY APPLICATIONS
K Sheelavant
IN Patent App. 202441093183 A , 2024
2024 - SUPERVISED LEARNING MODELS FOR STUDENT PERFORMANCE ANALYSIS WITH EXPLORATORY DATA ANALYSIS
K Sheelavant
IN Patent App. 202441071962 A , 2024
2024 - LOGISTIC REGRESSION AND RANDOM FOREST CLASSIFIER FOR ATTACK DETECTION IN IOT SENSOR DATA
K Sheelavant
IN Patent App. 202441064454 A , 2024
2024 - Pick Your Pet: Online Animal Trading Platform
S Gaddemane, R R, R H P, K Sheelavant
International Journal of All Research and Education Scientific Methods 10 (8 … , 2022
2022 - Correlative dynamic mapping based optimization (cdmo) for optimal allocation in cognitive radio sensor network
K Sheelavant, R Sumathi
Revista Geintec-Gestao Inovacao E Tecnologias 11 (4), 3955-3973 , 2021
2021
Citations: 4 - Dynamic Compilation of Pattern based clustering and Volumetric Probabilistic Mining for Network Routing in Cognitive Radio Sensor Networks
S Kumaresh, S R
Indian Journal of Science and Technology 14 (41), 3093-3106 , 2021
2021
Citations: 7 - Optimal Routing and Scheduling for Cognitive Radio Sensor Networks using Ensemble Multi Probabilistic Optimization and Truncated Energy Flow Classification Model
S Kumaresh, S R, KV Charan
International Journal of Engineering Trends and Technology 69 (9), 168-178 , 2021
2021
Citations: 9 - Interference and Delay Aware Routing Algorithm for Cognitive Radio Sensor Networks
K Sheelavant, R Sumathi
2017 2nd International Conference On Emerging Computation and Information … , 2017
2017
Citations: 1 - Energy efficient reliable routing through dynamic spectrum management in cognitive radio sensor networks
K Sheelavant, R Sumathi
2014 International Conference on Contemporary Computing and Informatics … , 2014
2014
Citations: 10
MOST CITED SCHOLAR PUBLICATIONS
- Ensemble Learning-Based Intrusion Detection and Classification for Securing IoT Networks: An Optimized Strategy for Threat Detection and Prevention
K Sheelavant
Journal of Intelligent Systems and Internet of Things 17 (2), 101-118 , 2025
2025
Citations: 39 - Energy efficient reliable routing through dynamic spectrum management in cognitive radio sensor networks
K Sheelavant, R Sumathi
2014 International Conference on Contemporary Computing and Informatics … , 2014
2014
Citations: 10 - Optimal Routing and Scheduling for Cognitive Radio Sensor Networks using Ensemble Multi Probabilistic Optimization and Truncated Energy Flow Classification Model
S Kumaresh, S R, KV Charan
International Journal of Engineering Trends and Technology 69 (9), 168-178 , 2021
2021
Citations: 9 - Fuzzy Temporal Rule-Based Natural Language Processig for Medical Text Analysis for Caring for Geriatric People
K Sheelavant, etc
International Journal of Basic and Applied Sciences 14 (Sl-1), 480-487 , 2025
2025
Citations: 7 - Dynamic Compilation of Pattern based clustering and Volumetric Probabilistic Mining for Network Routing in Cognitive Radio Sensor Networks
S Kumaresh, S R
Indian Journal of Science and Technology 14 (41), 3093-3106 , 2021
2021
Citations: 7 - Correlative dynamic mapping based optimization (cdmo) for optimal allocation in cognitive radio sensor network
K Sheelavant, R Sumathi
Revista Geintec-Gestao Inovacao E Tecnologias 11 (4), 3955-3973 , 2021
2021
Citations: 4 - Interference and Delay Aware Routing Algorithm for Cognitive Radio Sensor Networks
K Sheelavant, R Sumathi
2017 2nd International Conference On Emerging Computation and Information … , 2017
2017
Citations: 1 - Active Learning Pipelines for Efficient Annotation of Large-Scale IOT Video Streams
S Kumaresh, S Sanjay Kumar, S G, B L, JP J, S D
IEEE International Conference on Interdisciplinary Approaches in Technology … , 2026
2026 - Semantic Fact-Checking and Misinformation Analysis Using LLM and Generative AI
M T G, Jayshri, S Kumaresh, H K, SP Durga, M K
5th International Conference on Sentiment Analysis and Deep Learning (ICSADL … , 2026
2026 - A Neural Network Approach for Speech to Text Translation of Low Resource Language
K Sheelavant, P G P, S Shanbhag, S B C, S T, S Patil
1st International Conference on Advancement in Futuristic Technologies … , 2026
2026 - Deep Neural Network to Enhance Natural Image Structure for an Inverse Problem
K Sheelavant, TG Manjunath, A Prakash, DV Tejaswini, C S, D S
1st International Conference on Advancement in Futuristic Technologies … , 2026
2026 - Secure Object Identification Techniques for Autonomous Vehicles
K Sheelavant, SS Shirolkar, AS Mysore, SR Indu, S Shetty, N S K
1st International Conference on Advancement in Futuristic Technologies … , 2026
2026 - Improving Transparency and Adaptability in AI with Hybrid Generative Adversary Attention Networks
S Kumaresh, K Lakshmi Chandrakanth, S Nelli, A V, J L, H Poddar
Research Advances in Intelligent Computing 3, 396-410 , 2026
2026 - Real-Time Offline Edge AI Framework for Sensor Integrated Precision Agriculture
TG Manjunath, S Gowda, K Sheelavant, H Harika, I Sriram
2026 International Conference on AI-Driven Smart Systems and Ubiquitous … , 2026
2026 - DRA-Net with ISSA: A High-Performance Intrusion Detection Framework for IoT Networks with Enhanced Accuracy and Reduced False Alarms
P Mishra, TG Manjunath, AC Vikramathithan, K Sheelavant, N Nalinakshi, ...
2025 5th International Conference on Electrical, Computer and Energy … , 2025
2025 - Smart Gardening System with IoT Sensors and AI-Powered Plant Care
K Sheelavant
2025 - A SMART WEAPON DETECTION TECHNIQUE USING ARTIFICIAL INTELLIGENCE AND DEEP LEARNING FOR SECURITY APPLICATIONS
K Sheelavant
IN Patent App. 202441093183 A , 2024
2024 - SUPERVISED LEARNING MODELS FOR STUDENT PERFORMANCE ANALYSIS WITH EXPLORATORY DATA ANALYSIS
K Sheelavant
IN Patent App. 202441071962 A , 2024
2024 - LOGISTIC REGRESSION AND RANDOM FOREST CLASSIFIER FOR ATTACK DETECTION IN IOT SENSOR DATA
K Sheelavant
IN Patent App. 202441064454 A , 2024
2024 - Pick Your Pet: Online Animal Trading Platform
S Gaddemane, R R, R H P, K Sheelavant
International Journal of All Research and Education Scientific Methods 10 (8 … , 2022
2022
