Deep convolutional generative adversarial network algorithm based on improved fisher's criterion

Hao ZHANG; Guanglei QI; Xiaogang HOU; Kaimei ZHENG

doi:10.37188/OPE.20223024.3239

Optics and Precision Engineering, Volume. 30, Issue 24, 3239(2022)

Deep convolutional generative adversarial network algorithm based on improved fisher's criterion

Hao ZHANG1... Guanglei QI1,*, Xiaogang HOU2 and Kaimei ZHENG1 |Show fewer author(s)

Author Affiliations

¹Century College， Beijing University of Posts and Telecommunications， Beijing020， China

²College of Information Science and Engineering，Xinjiang University，Urumqi830046， China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. FDCGAN model architecture diagram

Download full size

View in Article

Fig. 2. Single stream network

Download full size

View in Article

Fig. 3. Shunt network

Download full size

View in Article

Fig. 4. Generating adversarial network model performance ratio （mean） on Fashion - MNIST dataset

Download full size

View in Article

Fig. 5. Model effect show picture

Download full size

View in Article

Fig. 6. Generalizability test on MNIST dataset

Download full size

View in Article

Fig. 7. Generalizability testing on the Fashion-MNIST dataset

Download full size

View in Article

Table 1. Quality of images generated by the model with different training samples

View table

View in Article

Table 1. Quality of images generated by the model with different training samples

Training Sample	Experimental method			Maximum Improvement
	DCGAN	FDCGAN	FDCGAN
	DCGAN	single-stream network	shunt network
10 000	99.79	91.76	95.17	8.03
20 000	92.25	92.14	92.42	0.11
30 000	88.39	83.57	83.15	5.24
40 000	80.13	79.72	77.38	2.75
50 000	78.08	73.84	78.81	4.24
60 000	75.42	74.73	73.70	1.72

Table 2. MNIST数据集上生成对抗网络模型性能比（）

View table

View in Article

Table 2. MNIST数据集上生成对抗网络模型性能比（）

Comparison Method	Number of training sessions					Improvement
Comparison Method	15	20	25	30	35	Improvement
AAE^［11］	96.34 $\pm$ 0.22	94.34 $\pm$ 1.31	95.71 $\pm$ 0.41	93.64 $\pm$ 1.35	92.50 $\pm$ 0.25	3.83
BGAN^［9］	203.88 $\pm$ 0.43	180.43 $\pm$ 1.08	168.45 $\pm$ 1.61	161.91 $\pm$ 1.05	136.67 $\pm$ 0.72	67.21
LSGAN^［6］	90.60 $\pm$ 0.20	90.61 $\pm$ 0.12	84.30 $\pm$ 0.20	83.59 $\pm$ 1.10	82.26 $\pm$ 0.63	8.34
ACGAN^［12］	86.48 $\pm$ 0.41	84.78 $\pm$ 0.95	80.19 $\pm$ 1.15	76.90 $\pm$ 0.89	73.92 $\pm$ 0.60	12.55
WGAN^［7］	152.60 $\pm$ 2.35	149.58 $\pm$ 0.01	125.97 $\pm$ 0.05	120.23 $\pm$ 0.39	124.03 $\pm$ 1.03	28.57
DCGAN^［10］	85.47 $\pm$ 0.57	81.68 $\pm$ 0.44	84.06 $\pm$ 0.78	80.71 $\pm$ 0.98	81.77 $\pm$ 0.65	3.7
FDCGAN single-stream network	81.21 $\pm$ 0.38	77.51 $\pm$ 0.43	74.73 $\pm$ 1.39	76.05 $\pm$ 1.01	75.94 $\pm$ 0.58	5.27
FDCGAN shunt network	80.36 $\pm$ 0.21	78.72 $\pm$ 1.16	73.70 $\pm$ 1.24	72.16 $\pm$ 0.58	73.20 $\pm$ 0.01	7.16

Tools

Get Citation

Copy Citation Text

Hao ZHANG, Guanglei QI, Xiaogang HOU, Kaimei ZHENG. Deep convolutional generative adversarial network algorithm based on improved fisher's criterion[J]. Optics and Precision Engineering, 2022, 30(24): 3239

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites