Edge-Optimized Vision Transformer Architecture for Ultra-Low-Latency Object Detection in 6G IoT Devices

View at publisher Summarise with VIRAT
Authors: Anusha P, Yousuf Hazaimeh, T. Vijetha, Geetha R, Akila Venkatraman, Ramya M
Cited by: 0
References: 16

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The advancements in the high demand for ultra-low-latency intelligence in large-scale IoT networks have enhanced the demands of edge-deployable vision models that can detect objects in real-time. In order to satisfy this need, an Edge-Optimized Vision Transformer design is created based on spectral token compression, entropy-directed patch refinement, latency-constrained multi-head attention, as well as hierarchical feature fusion that is optimized to run on limited edge devices. The model is trained with a balanced and high-variability object detection data set, which ensures that it can be resistant to illumination variations, motion blur, and small-object cases. The proposed system is experimentally found to have 96.8 % detection accuracy,$94.7 \% \text{mAP} {@} 0.5$, and 4.3 ms end-to-end latency on edge-class IoT hardware. Also, the architecture can save 41 % FLOPs of computational loads, 38 % of the energy consumption, and 22 % of the small-object recall of lightweight transformer baselines. The findings verify that the proposed

Fields of Study & Subjects

Advanced Neural Network Applications IoT and Edge/Fog Computing Visual Attention and Saliency Detection

Publication details

Journal / source
International Conference on Power Electronics & IoT Applications in Renewable Energy and its Control (PARC)
Pages
228–233
Publication date
20 February 2026
Language
EN