loading page

SAGP: a spectral attention-based global pooling
  • +2
  • Xiang Yu,
  • Longzheng Xu,
  • Yi Han,
  • Zhe Geng,
  • Daiyin Zhu
Xiang Yu
Nanjing Institute of Technology

Corresponding Author:[email protected]

Author Profile
Longzheng Xu
Nanjing Institute of Technology
Author Profile
Yi Han
Nanjing Institute of Technology
Author Profile
Zhe Geng
Nanjing University of Aeronautics and Astronautics
Author Profile
Daiyin Zhu
Nanjing University of Aeronautics and Astronautics College of Electronic and Information Engineering
Author Profile

Abstract

Pooling operations, essential for neural networks (NNs), reduce feature map dimensions while preserving key features and enhancing spatial invariance. Traditional pooling methods often miss the feature maps’ alternating current (AC) components. This study introduces a novel global pooling technique utilizing spectral self-attention, leveraging the discrete cosine transform (DCT) for spectral analysis and a self-attention mechanism for assessing frequency component significance. This approach allows for efficient feature synthesis through weighted averaging, significantly boosting TOP-1 accuracy with minimal parameter increase, outperforming existing models.
Submitted to Electronics Letters
12 Apr 2024Submission Checks Completed
12 Apr 2024Assigned to Editor
12 Apr 2024Review(s) Completed, Editorial Evaluation Pending
17 Apr 2024Reviewer(s) Assigned
16 May 2024Editorial Decision: Revise Major
29 May 20241st Revision Received
29 May 2024Submission Checks Completed
29 May 2024Assigned to Editor
29 May 2024Review(s) Completed, Editorial Evaluation Pending