Multi-Head-Self-Attention based YOLOv5X-transformer for multi-scale object detection

The state-of-the-art deep learning models mostly depend upon the region proposal and grid methods in detecting the objects with localization has been in practice for a long time but still, it has got scope for further improvement. There exists a visual challenge of small-scale object detection. To s...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia tools and applications 2024-04, Vol.83 (12), p.36491-36517
Main Authors: Vasanthi, Ponduri, Mohan, Laavanya
Format: Article
Language:English
Subjects:
1230: Sentient Multimedia Systems and Visual Intelligence
Aliasing
Computer Communication Networks
Computer Science
Data Structures and Information Theory
Deep learning
Feature maps
Image enhancement
Image quality
Multimedia
Multimedia Information Systems
Object recognition
Proposals
Semantics
Spatial data
Special Purpose and Application-Based Systems
Transformers
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The state-of-the-art deep learning models mostly depend upon the region proposal and grid methods in detecting the objects with localization has been in practice for a long time but still, it has got scope for further improvement. There exists a visual challenge of small-scale object detection. To solve this issue, some methods are used for creating and enhancing shallow features and feature fusion concepts. However, the progressive enhancement of shallow features reduces the quality of the image, and feature fusion causes the aliasing effect. To ensure and solve this aliasing effect in small-scale object detection, this paper introduces the YOLOv5X- transformer. In this model, Multi-Head-Self-Attention (MHSA) module extracts in-depth information from the feature maps based on the query, key, and value parameters. Afterward, these feature maps are pooled together at five different scales by using Spatial Pyramid Pooling-Faster (SPPF) to improve the quality of the feature maps. To save spatial information and locate the pixels correctly Path-Aggregated Network (PANet) is used as a neck model. This model is experimentally verified on the PASCAL dataset. This model has achieved 87.7% mAP, 85.2% Precision, and 81.4% Recall. These tested results show that the proposed model performs better than existing models in detecting small objects.
ISSN:1573-7721
1380-7501
1573-7721
DOI:10.1007/s11042-023-15773-4