Enhanced mechanisms of pooling and channel attention for deep learning feature maps

被引:2
作者
Li, Hengyi [1 ]
Yue, Xuebin [1 ]
Meng, Lin [2 ]
机构
[1] Ritsumeikan Univ, Grad Sch Sci & Engn, Kusatsu, Shiga, Japan
[2] Ritsumeikan Univ, Coll Sci & Engn, Kusatsu, Shiga, Japan
关键词
DNNs; Max pooling; Average pooling; FMAPooling; Self-attention; FMAttn;
D O I
10.7717/peerj-cs.1161
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
The pooling function is vital for deep neural networks (DNNs). The operation is to generalize the representation of feature maps and progressively cut down the spatial size of feature maps to optimize the computing consumption of the network. Furthermore, the function is also the basis for the computer vision attention mechanism. However, as a matter of fact, pooling is a down-sampling operation, which makes the feature-map representation approximately to small translations with the summary statistic of adjacent pixels. As a result, the function inevitably leads to information loss more or less. In this article, we propose a fused max-average pooling (FMAPooling) operation as well as an improved channel attention mechanism (FMAttn) by utilizing the two pooling functions to enhance the feature representation for DNNs. Basically, the methods are to enhance multiple-level features extracted by max pooling and average pooling respectively. The effectiveness of the proposals is verified with VGG, ResNet, and MobileNetV2 architectures on CIFAR10/100 and ImageNet100. According to the experimental results, the FMAPooling brings up to 1.63% accuracy improvement compared with the baseline model; the FMAttn achieves up to 2.21% accuracy improvement compared with the previous channel attention mechanism. Furthermore, the proposals are extensible and could be embedded into various DNN models easily, or take the place of certain structures of DNNs. The computation burden introduced by the proposals is negligible.
引用
收藏
页数:18
相关论文
共 50 条
[21]   RoseSegNet: An attention-based deep learning architecture for organ segmentation of plants [J].
Turgut, Kaya ;
Dutagaci, Helin ;
Rousseau, David .
BIOSYSTEMS ENGINEERING, 2022, 221 :138-153
[22]   6G Technology for Indoor Localization by Deep Learning with Attention Mechanism [J].
Chiu, Chien-Ching ;
Wu, Hung-Yu ;
Chen, Po-Hsiang ;
Chao, Chen-En ;
Lim, Eng Hock .
APPLIED SCIENCES-BASEL, 2024, 14 (22)
[23]   Self-Attention-Based Deep Learning Network for Regional Influenza Forecasting [J].
Jung, Seungwon ;
Moon, Jaeuk ;
Park, Sungwoo ;
Hwang, Eenjun .
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, 2022, 26 (02) :922-933
[24]   An Enhanced Deep Knowledge Tracing Model via Multiband Attention and Quantized Question Embedding [J].
Xu, Jiazhen ;
Hu, Wanting .
APPLIED SCIENCES-BASEL, 2024, 14 (08)
[25]   Stripe-Transformer: deep stripe feature learning for music source separation [J].
Jiale Qian ;
Xinlu Liu ;
Yi Yu ;
Wei Li .
EURASIP Journal on Audio, Speech, and Music Processing, 2023
[26]   Stripe-Transformer: deep stripe feature learning for music source separation [J].
Qian, Jiale ;
Liu, Xinlu ;
Yu, Yi ;
Li, Wei .
EURASIP JOURNAL ON AUDIO SPEECH AND MUSIC PROCESSING, 2023, 2023 (01)
[27]   EBiDA-FPN: enhanced bi-directional attention feature pyramid network for object detection [J].
Yang, Xiaobao ;
He, Yulong ;
Wu, Junsheng ;
Wang, Wentao ;
Sun, Wei ;
Ma, Sugang ;
Hou, Zhiqiang .
JOURNAL OF ELECTRONIC IMAGING, 2024, 33 (02)
[28]   Effective person re-identification by self-attention model guided feature learning [J].
Li, Yang ;
Jiang, Xiaoyan ;
Hwang, Jenq-Neng .
KNOWLEDGE-BASED SYSTEMS, 2020, 187 (187)
[29]   Convolutional-Attention Hybrid Deep-Learning Model With Transfer Learning for Quality Inspection of DC Motors [J].
Xie, Wei ;
Li, Yanfu ;
Wei, Haixiang ;
Zhang, Langwen .
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, 2024, 73 :1-13
[30]   A Self-Attention-Based Deep Reinforcement Learning Approach for AGV Dispatching Systems [J].
Wei, Qinglai ;
Yan, Yutian ;
Zhang, Jie ;
Xiao, Jun ;
Wang, Cong .
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2024, 35 (06) :7911-7922