Multi Strategy Parrot Optimized Convolutional Vision Transformer Network for High Precision Image Recognition
Data and Metadata · 12.03.2026 · 5 · 793
DOI: 10.56294/dm2026793 ↗Kurz gefasst
Fachartikel in der Zeitschrift Data and Metadata, online erschienen am 12.03.2026. Verfasst von Dongshuai Zhang. Die Arbeit nennt 20 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Introduction
High-accuracy image recognition remains an urgent challenge because it requires simultaneous extraction of fine-grained local features and complex global visual characteristics. Traditional Convolutional Neural Networks (CNNs) effectively learn local textures and edges but are often ineffective in modeling large-scale dependencies, Vision Transformers (ViTs) provide strong global context modeling. However, they usually demand high computational resources and sensitive parameter tuning.
Objective
The objective of this work is to develop an optimized hybrid image recognition framework that effectively captures and fuses local and global features while improving convergence behavior and classification precision.
Method
A Multi-Strategy Parrot enhanced Convolutional Vision Transformer Network (MSP-CVTN) is proposed by integrating CNN and ViT models with a metaheuristic optimization strategy. Experiments are conducted on the fine-grained CUB-200-2011 bird species classification dataset. Image preprocessing includes Min-max normalization and histogram equalization, followed by Gabor filter-based texture and pattern feature extraction to enhance image representations. The CNN component captures local fine-scale features, while the ViT component models global dependencies between feature tokens. Multi-Strategy Parrot Optimization (MSPO) regulates CNN filter parameters, Transformer attention weights, and feature fusion strategies to enhance convergence and classification performance. Model implementation is carried out using Python with TensorFlow and PyTorch frameworks.
Results
Quantitative experimental results demonstrate that the proposed MSP-CVTN achieves an accuracy of 0,9823, significantly outperforming classical deep learning methods. The optimized feature fusion and parameter tuning contribute to improved convergence stability and high-precision recognition performance.
Conclusions
The proposed MSP-CVTN provides a powerful and flexible framework for complex image recognition tasks by effectively combining local and global feature learning. Its optimized architecture delivers superior accuracy and could be adapted to a wide range of image recognition datasets.
Zitieren
APA
Zhang, D. (2026). Multi Strategy Parrot Optimized Convolutional Vision Transformer Network for High Precision Image Recognition. Data and Metadata, 5, 793. https://doi.org/10.56294/dm2026793
Vancouver
Zhang D. Multi Strategy Parrot Optimized Convolutional Vision Transformer Network for High Precision Image Recognition. Data and Metadata. 2026;5:793. doi:10.56294/dm2026793
BibTeX
@article{Zhang2026,
author = {Zhang, Dongshuai},
title = {Multi Strategy Parrot Optimized Convolutional Vision Transformer Network for High Precision Image Recognition},
journal = {Data and Metadata},
year = {2026},
volume = {5},
pages = {793},
doi = {10.56294/dm2026793}
}
EUVIMED