Open Access

Fast fourier transform based new pooling layer for deep learning

Aqeel Mohsin Hamad
Aqeel Mohsin Hamad
   | Apr 16, 2022
International Journal on Smart Sensing and Intelligent Systems's Cover Image
About this article
Previous Article
Next Article
Cite
Published Online: Apr 16, 2022
Page range: 1 - 14
Received: Jul 01, 2021
DOI: https://doi.org/10.21307/ijssis-2022-0003
Keywords
© 2022 Aqeel Mohsin Hamad et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1:

Proposed pooling layer.
Proposed pooling layer.

Figure 2:

Proposed first (FTM1) algorithm.
Proposed first (FTM1) algorithm.

Figure 3:

Proposed second (FTM2) algorithm.
Proposed second (FTM2) algorithm.

Figure 4:

Proposed third (FTM3) algorithm.
Proposed third (FTM3) algorithm.

Figure 5:

Original images, pooled images and reconstructed images by FTM1.
Original images, pooled images and reconstructed images by FTM1.

Figure 6:

Accuracy of MNIST classification for proposed method.
Accuracy of MNIST classification for proposed method.

Figure 7:

Sensitivity results for proposed method for MNIST classification.
Sensitivity results for proposed method for MNIST classification.

Figure 8:

Specificity results for proposed method for MNIST classification.
Specificity results for proposed method for MNIST classification.

Figure 9:

Training steps for MNIST classification by (FTM1) method.
Training steps for MNIST classification by (FTM1) method.

Figure 10:

Accuracy results for proposed method for CIFAR 10 classification.
Accuracy results for proposed method for CIFAR 10 classification.

Figure 11:

Specificity results for proposed method for CIFAR 10 classification.
Specificity results for proposed method for CIFAR 10 classification.

Figure 12:

Precision results for proposed method for CIFAR 10 classification.
Precision results for proposed method for CIFAR 10 classification.

Proposed methods results for MNIST dataset.

Method Saeedan et al. (2018) Lee et al. (2016) FTM1 FTM2 FTM3
Accuracy (%) 98.80 98.72 99.95 99.84 99.96

Accuracy results for CIFAR10 dataset classification.

Method Saeedan et al. (2018) Lee et al. (2016) RFT1 RFT2 RFT3
Accuracy (%) 72.59 72.4 73.88 73.82 73.76

Performance of proposed methods.

Images Metrics Saeedan et al. (2018) Lee et al. (2016) FTM1 FTM2 FTM3
Lena SNR dB 23.28 24.01 24.1754 24.1603 24.16785
SSIM 0.7882 0.7893 0.79396 0.77886 0.78641
Correlation 0.9822 0.9833 0.9846 0.9695 0.97705
Cameraman SNR 20.14 21.87 20.2935 20.2784 20.28595
SSIM 0.7854 0.7867 0.7891 0.774 0.78155
Correlation 0.9814 0.9843 0.9952 0.9801 0.98765
Barbara SNR 23.69 22.13 27.336 27.3209 27.32845
SSIM 0.7041 0.7065 0.7092 0.6941 0.70165
Correlation 0.9622 0.9648 0.9612 0.9461 0.95365
Test image SNR 28.98 28.78 29.38 29.3649 29.37245
SSIM 0.7832 0.7841 0.7852 0.7701 0.77765
Correlation 0.9913 0.9922 0.9965 0.9814 0.98895

Confusion matrix for (FTM1) method for MNIST Classification.

Target class
Class 0 Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7 Class 8 Class 9 Sensitivity
Output class
 Class 0 250 0 0 0 0 0 0 0 0 0 100
 Class 1 0 250 0 0 0 0 0 0 0 0 100
10% %
 Class 2 0 0 250 0 0 0 0 0 0 0 100
10% %
 Class 3 0 0 0 250 0 0 0 0 0 0 100
10% %
 Class 4 0 0 0 0 250 0 0 0 0 0 100
10% %
 Class 5 0 0 0 0 0 250 0 0 0 0 100
10% %
 Class 6 0 0 0 0 0 0 249 0 0 0 100
10% %
 Class 7 0 0 0 0 0 0 0 250 0 0 100
10% %
 Class 8 0 0 0 0 0 0 1 0 250 0 99.9
10% 6%
 Class 9 0 0 0 0 0 0 0 0 0 250 100
10% %
100 100 100 100 100 100 100 99.9 100 100 100
% % % % % % % 6% % % %

Confusion matrix for (FTM1) for CIFAR.10 classification.

Target class
airplane auto mobi bird cat deer dog frog hour ship truc Sens
Output class
 airpla 796 29 68 32 25 13 10 22 57 3 73
 autom 10 829 6 6 4 2 3 4 15 7 86
 bird 44 10 599 62 69 42 37 32 13 8 66
 cat 18 3 61 533 36 16 57 43 10 9 56
 deer 24 3 75 60 708 38 26 55 5 1 71
 dog 4 4 109 201 55 69 22 97 4 3 58
 frog 9 9 46 66 54 13 83 5 7 1 78
 hours 7 1 16 12 38 25 3 72 2 4 86
 ship 48 36 9 14 9 6 4 6 86 2 82
 truck 38 75 8 13 4 6 2 8 18 80 83
 specifi 79.8 82.3 59.8 53.3 71.6 69 83 73 86 82 74
eISSN:
1178-5608
Language:
English
Publication timeframe:
Volume Open
Journal Subjects:
Engineering, Introductions and Overviews, other
Journal RSS Feed