Low-parameter hybrid attention model based image classification

Lan Lin; Shisong Tan; Feifei Long

doi:10.1117/12.2660001

30 November 2022 Low-parameter hybrid attention model based image classification

Lan Lin, Shisong Tan, Feifei Long

Proceedings Volume 12456, International Conference on Artificial Intelligence and Intelligent Information Processing (AIIIP 2022); 124562H (2022) https://doi.org/10.1117/12.2660001
Event: International Conference on Artificial Intelligence and Intelligent Information Processing (AIIIP 2022), 2022, Qingdao, China

Abstract

The hybrid domain attention model was introduced into deep learning network and has good performance in computer vision recognition and classification tasks for considering both spatial and channel domain information, which leads to higher network complicity, more parameters and much training time. In this paper, an Efficient Hybrid Attention(EHA) model was proposed. The spatial and channel domain information was extracted and fused into the EHA-block, that can be embedded in deep network Resnet50 and lightweight mobile network MobileNetV2 to improve their capabilities with low parameter- increment. The experimental results on CIFAR10&100 and miniImageNet show that the number of parameters in EHA model decreases by 15.4% compared with CA hybrid model, and only increases by 0.49% compared with ECA channel model. Moreover, the accuracy of EHA classification is also improved.

Citation Download Citation

Lan Lin, Shisong Tan, and Feifei Long "Low-parameter hybrid attention model based image classification", Proc. SPIE 12456, International Conference on Artificial Intelligence and Intelligent Information Processing (AIIIP 2022), 124562H (30 November 2022); https://doi.org/10.1117/12.2660001

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available