Lightweight Image Classification Network Based on Feature Extraction Network SimpleResUNet and Attention
Primary Area: representation learning for computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: CNN, ResNet, Attention, Feature Space, Lightweight Neural Network
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: A novel Residual U-net network architecture is proposed, which effectively preserves the advantages of ResNet and U-net while incorporating a Self-Attention classifier.
Abstract: Using lightweight neural network models for small-sample image classification tasks has always been a challenging task. This paper proposes a feature extraction network called SimpleResUNet based on ResNet and U-Net, and adopts the Attention mechanism as the feature classifier for image classification, aiming to improve the accuracy and robustness of small-sample image classification tasks using lightweight network structures. Firstly, the network combines the feature extraction capability of U-Net and the efficient feature propagation capability of ResNet to effectively extract details and contextual information from the images. Secondly, the Attention mechanism is used to capture the correlations and dependencies between different features in the feature sequence. Multiple public datasets were used for verification in the experiment, and comparative analysis was conducted with other methods. Experimental results show that the network achieves superior performance in image classification tasks. Finally, some thoughts on the mechanism of this model are discussed, the work of this paper is summarized, and future research directions are prospected.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 36
Loading