Open to R&D collaboration · Mysuru, India

Research that helps machines see the road clearly.

Somanna M — AI & Computer Vision Researcher · AI Programme Manager, Planet Globe

I work on deep learning for object detection and recognition: designing lightweight architectures, building large-scale datasets, and leading applied AI programmes from experiment to production.

Email me → LinkedIn GitHub

Ph.D. thesis submittedUniversity of Mysore

Scopus & Web of ScienceSpringer · IEEE · Elsevier

Visiting ProfessorHuanghuai University, China

AI AI Programme ManagerPlanet Globe · Mysuru

Research

A research program for robust, efficient road perception

Doctoral work at the University of Mysore: novel architectures, a large-scale dataset, and the training methodology to make traffic-sign detection dependable in the real world.

Doctoral research supervisor Prof. H. S. Nagendraswamy Department of Studies in Computer Science, University of Mysore Professor of Computer Science with over 25 years in academia and research spanning pattern recognition, image processing, fuzzy theory, and symbolic data analysis, with publications in venues such as Pattern Recognition Letters and Neurocomputing. Reviewer for Pattern Recognition and Pattern Recognition Letters, program chair of international conferences including DAL 2018 and ICCR 2011, and Chairman of the Board of Studies in Computer Science, University of Mysore.

View profile →

Architecture · Classification

ReSHOGNet

Hybrid multi-scale residual CNN fused with DiffHOG features for lightweight, efficient traffic-sign classification.

Architecture · Transformer

LHCViT

Lightweight hybrid convolutional vision transformer with multi-scale feature fusion and hierarchical positional encoding.

Detection · Small objects

FSH-NSSA-YOLO

Four-scale hybrid YOLO with non-semantic sparse attention for detecting small traffic signs in complex scenes.

Dataset · Benchmark

NRTSD

Large-scale natural real-time Indian traffic-sign dataset with weather augmentation for detection and recognition.

Methodology

Training techniques

Iterative training and augmentation for higher detection efficiency on Indian roads.

Data

Data-centric methods

Dataset curation and augmentation pipelines, including GAN-based augmentation.

Evaluation

Empirical studies

Comparative analysis of weight initialization and optimizers on classification performance.

Collaboration

Applied vision

Signature and document element detection, segmentation, and image forgery forensics.

Open-source · PyPI

DeepLabelNet

Python package for AI-assisted image annotation and dataset development, published on PyPI.

View on PyPI →

pip install deeplabelnet

Publications

Peer-reviewed research

Springer LNNS, IEEE and Elsevier venues, indexed in Scopus and Web of Science. Filter by type below.

01

ReSHOGNet: A Novel Hybrid Multi-Scale Residual CNN with DiffHOG Features for Lightweight and Efficient Traffic Sign Classification

Somanna M, H. S. Nagendraswamy, Thejashwini B L, Manjuchethan N, Chethana Kumara B M

Indian Journal of Science and Technology, 19(10), 736–750 · Web of Science

doi:10.17485/IJST/v19i10.303

2026published
02

Improving Document Element Detection Through Single and Two Stage Model Integration

Thejashwini B L, H. S. Nagendraswamy, Keerthana H, Somanna M

Indian Journal of Science and Technology, 19(16) · Web of Science

doi:10.17485/IJST/v19i16.304

2026published
03

A Comparative Analysis of Deep Learning Weight/Kernel Initialization and Optimization Techniques on Traffic Sign Classification Performance

Somanna M, H. S. Nagendraswamy, Amreen Sultana, Thejashwini B L, Chethana Kumara B M

Innovative Computing and Communications (ICICC 2026), LNNS 2045, Springer, 1–16 · Scopus

doi:10.1007/978-3-032-30310-3_35

2026published
04

Integrating Handcrafted Features and Deep Learning Models for Image Splicing Detection: A Hybrid Feature Fusion Approach

Rajashekara M, H. K. Chethan, Thejashwini B L, Somanna M

Indian Journal of Science and Technology · Web of Science

2026accepted
05

Optimized Keypoint-Based Copy-Move Forgery Detection

Rajashekara M, H. K. Chethan, Pavan Kumar B S, Thejashwini B L, Somanna M

ComSIA 2026, Springer LNNS · Scopus, Web of Science, EI, DBLP

2026accepted
06

LHCViT: A Lightweight Hybrid Convolutional Vision Transformer with Multi-Scale Feature Fusion and Hierarchical Positional Encoding for Traffic Sign Classification

Somanna M, H. S. Nagendraswamy, Thejashwini B L, Manjuchethan N, Rajashekara M, Chethana Kumara B M

Intelligent Automation & Soft Computing

2026communicated
07

FSH-NSSA-YOLO: A Four-Scale Hybrid YOLO with Non-Semantic Sparse Attention for Small Traffic-Sign Detection

H. S. Nagendraswamy, D. S. Guru, Somanna M

MethodsX, Elsevier · Scopus, Web of Science

2026communicated
08

NRTSD: A Large-Scale Natural Real-Time Indian Traffic Sign Dataset with Weather Augmentation for Detection and Recognition

H. S. Nagendraswamy, D. S. Guru, Somanna M, Manjuchethan N, Thejashwini B L

IEEE Data Descriptions, IEEE

2026communicated
09

Enhancing Indian Traffic Sign Detection Efficiency with Iterative Training Techniques and Augmentation

Somanna M, H. S. Nagendraswamy, Jyothi V K, Preethi N

Data Analytics and Learning (DAL 2024), LNNS 1540, Springer, 267–284 · Scopus

doi:10.1007/978-981-96-9533-1_19

2025published
10

Enhancing Classification Accuracy Through a Cluster-Based Training Approach: A Case Study of Butterfly Species Classification

Manjuchethan N, H. S. Nagendraswamy, Somanna M, Kumara C

IEEE ISACC 2025, 1007–1012

doi:10.1109/ISACC65211.2025.10969317

2025published
11

Enhanced Butterfly Image Segmentation Using Deep Encoder–Decoder Based Architecture

Manjuchethan N, H. S. Nagendraswamy, Siddappa Revappa Mudhale, Thejashwini B L, Somanna M

4th International Conference on Machine Learning and Data Engineering (ICMLDE 2025)

2025accepted
12

Precise Forgery Region Identification in Digital Images: A Novel Approach Integrating SIFT, DBSCAN, and K-Means

Rajashekara M, H. K. Chethan, Somanna M, Thejashwini B L

International Journal of Research and Analysis in Science and Engineering, 4(5)

2024published
13

Automated Template Based Signature Detection and Extraction from Multilingual Documents

Thejashwini B L, H. S. Nagendraswamy, Supritha B, Somanna M

IEEE CVMI 2024, 1–6

doi:10.1109/CVMI61877.2024.10782096

2024published
14

Enhancing Signature Detection from Scanned Documents Using Hybrid Detection Model and Post Processing

Thejashwini B L, H. S. Nagendraswamy, Rakshitha P, Somanna M

Third International Conference on Data Analytics and Learning (DAL 2024)

doi:10.1007/978-981-96-9533-1_18

2024published
15

Forgery Detection in Digital Images Using Transfer Learning Approaches

Rajashekara M, H. K. Chethan, Thejashwini B L, Somanna M, Manjuchethan N

ICRRACS 2023 · ISBN 978-81-967420-6-5

2023published

Experience

Industry, research and teaching

Sept 2025 – Present
AI Programme Manager

Planet Globe (Paranjyothi Enterprises Pvt. Ltd.) · Mysuru, India

Leading AI initiatives across computer vision, AI-based data annotation, and GIS applications at an ISO 27001:2022 and ISO 9001:2015 certified technology company. Directing enterprise platform development, including the HRVOLVE attendance and leave management system, and technical preparation of GIS e-governance tender submissions.
Mar – May 2026
Visiting Professor

Huanghuai University · Zhumadian, Henan, China

Taught Introduction to Programming II and Mobile Application Development to undergraduate students. Teaching recognized with appreciation from students and teaching staff of the university.
2021 – Present
Doctoral Researcher

Department of Studies in Computer Science, University of Mysore · thesis submitted

Deep learning and computer vision for robust, efficient traffic sign detection and recognition in autonomous driving scenarios: novel lightweight architectures, iterative training and GAN-based augmentation, large-scale dataset development, and systematic empirical evaluation.
2019 – 2022
Software Developer

Q Ventures Pvt. Ltd.

Engineered enterprise automation and e-governance platforms for educational institutions: an admission management system processing 10,000+ applications annually, secure payment gateway integration with a 99.5% transaction success rate, document management with role-based access, and an international online admission portal for Laguna State Polytechnic University, Philippines.

Education

2021 –
Ph.D. in Computer Science · thesis submitted
University of Mysore. Deep learning and computer vision for traffic sign detection and recognition.
2023
UGC-NET, qualified
Computer Science and Applications
2017
KSET, qualified
Computer Science and Applications
2017
MCA · Third Rank
University of Mysore, Manasagangotri. GPA 8.972
2014
B.Sc. Physics, Mathematics, Computer Science
Yuvaraja's College, University of Mysore. Highest Computer Science score in the batch (84.8%)

Skills

AI / ML & Deep Learning

PyTorchTensorFlowKerasScikit-LearnCNNsVision TransformersYOLO

Computer Vision

OpenCVMATLAB

Programming

PythonJavaGroovyCJavaScriptTypeScriptPHP

Frameworks & Web

GrailsSpringDjangoNode.js.NETAngularReactREST APIsGraphQL

Databases · Cloud · Tools

MySQLPostgreSQLMSSQLMongoDBAWSGCPDockerKubernetesGitLaTeX

A research program for robust, efficient road perception

ReSHOGNet

LHCViT

FSH-NSSA-YOLO

NRTSD

Training techniques

Data-centric methods

Empirical studies

Applied vision

DeepLabelNet

Peer-reviewed research

ReSHOGNet: A Novel Hybrid Multi-Scale Residual CNN with DiffHOG Features for Lightweight and Efficient Traffic Sign Classification

Improving Document Element Detection Through Single and Two Stage Model Integration

A Comparative Analysis of Deep Learning Weight/Kernel Initialization and Optimization Techniques on Traffic Sign Classification Performance

Integrating Handcrafted Features and Deep Learning Models for Image Splicing Detection: A Hybrid Feature Fusion Approach

Optimized Keypoint-Based Copy-Move Forgery Detection

LHCViT: A Lightweight Hybrid Convolutional Vision Transformer with Multi-Scale Feature Fusion and Hierarchical Positional Encoding for Traffic Sign Classification

FSH-NSSA-YOLO: A Four-Scale Hybrid YOLO with Non-Semantic Sparse Attention for Small Traffic-Sign Detection

NRTSD: A Large-Scale Natural Real-Time Indian Traffic Sign Dataset with Weather Augmentation for Detection and Recognition

Enhancing Indian Traffic Sign Detection Efficiency with Iterative Training Techniques and Augmentation

Enhancing Classification Accuracy Through a Cluster-Based Training Approach: A Case Study of Butterfly Species Classification

Enhanced Butterfly Image Segmentation Using Deep Encoder–Decoder Based Architecture

Precise Forgery Region Identification in Digital Images: A Novel Approach Integrating SIFT, DBSCAN, and K-Means

Automated Template Based Signature Detection and Extraction from Multilingual Documents

Enhancing Signature Detection from Scanned Documents Using Hybrid Detection Model and Post Processing

Forgery Detection in Digital Images Using Transfer Learning Approaches

Industry, research and teaching

AI Programme Manager

Visiting Professor

Doctoral Researcher

Software Developer

Education

Ph.D. in Computer Science · thesis submitted

UGC-NET, qualified

KSET, qualified

MCA · Third Rank