Super-Resolution Computed Tomography Reconstruction of Residual Attention Aggregation Dual Regression Network

Attentional mechanism		Activation function	Params /M	PSNR/ dB	SSIM
Channel attention	CA^［13］	Sigmoid（·）	4.80	37.303	0.9192
	CoA^［16］	h_swish（·）	4.82	37.105	0.9186
	ECA^［15］	Sigmoid（·）	4.77	37.396	0.9196
	CLA	Sigmoid（·）	4.77	37.679	0.9202
	CLA	1+tanh（·）	4.77	37.746	0.9205
Spatial attention	SAM^［17］	Sigmoid（·）	4.78	36.374	0.9127
	SLA	Sigmoid（·）	4.77	36.827	0.9141
	SLA	1+tanh（·）	4.77	36.954	0.9148

Table 3. Ablation test results of algorithm
View table
Table 3. Ablation test results of algorithm
SDEB MFDEB CRCAB CRCLAB RAAB Params /M PSNR /dB SSIM
√ √ 4.80 37.303 0.9192
√ √ 4.81 37.549 0.9197
√ √ 4.77 37.746 0.9205
√ √ 4.78 37.787 0.9207
√ √ 4.79 37.770 0.9210
√ √ 4.80 37.834 0.9210

Table 4. Weight factor $λ$ configuration
View table
Table 4. Weight factor $λ$ configuration
$λ$ 0 0.01 0.05 0.1 0.5 1.0
PSNR 37.662 37.797 37.818 37.834 37.636 37.531

Table 5. Analysis of reconstruct time for SRX4
View table
Table 5. Analysis of reconstruct time for SRX4
Dataset Size /pixcel Number DRN /s RAADRNet /s Time increment /s
Set1 1024 $\times$ 1024 20 3.01 3.13 0.12
Set2 512 $\times$ 512 20 0.30 0.32 0.02
Set3 512 $\times$ 512 20 0.53 0.57 0.04

Table 6. Data of experimental results

View table

Table 6. Data of experimental results

Algorithm	EI	Set1			Set2			Set3
Algorithm	EI	X2	X4	X16	X2	X4	X8	X2	X4	X8
Bicubic	PSNR /dB	34.470	31.472	27.450	27.346	24.049	19.891	31.554	27.475	22.547
Bicubic	SSIM	0.9008	0.8616	0.8141	0.9083	0.8233	0.7094	0.9313	0.8550	0.7401
SRCNN	PSNR /dB	35.339	31.940	28.000	29.296	25.780	21.488	32.783	30.317	25.071
SRCNN	SSIM	0.9062	0.8714	0.8230	0.9269	0.8588	0.7739	0.9397	0.8940	0.8098
RDN	PSNR /dB	36.205	32.270	28.362	30.031	26.526	22.029	33.467	31.156	25.921
RDN	SSIM	0.9095	0.8742	0.8272	0.9361	0.8703	0.7905	0.9442	0.9034	0.8039
RCAN	PSNR /dB	36.705	32.681	28.635	30.999	27.283	22.573	34.098	31.792	26.873
RCAN	SSIM	0.9161	0.8780	0.8283	0.9391	0.8788	0.8048	0.9471	0.9087	0.8423
SAN	PSNR /dB	37.008	33.041	28.866	31.743	27.832	22.935	34.601	32.223	27.347
SAN	SSIM	0.9183	0.8818	0.8376	0.9420	0.8852	0.8122	0.9503	0.9091	0.8492
DRN	PSNR /dB	37.303	33.377	29.121	32.675	28.209	23.285	34.964	32.820	27.928
DRN	SSIM	0.9192	0.8833	0.8414	0.9473	0.8885	0.8184	0.9541	0.9141	0.8566
RAADRNet	PSNR /dB	37.834	34.176	29.483	33.402	28.449	23.544	35.434	33.480	28.312
RAADRNet	SSIM	0.9210	0.8866	0.8440	0.9498	0.8912	0.8219	0.9547	0.9176	0.8609

Tools

Get Citation

Copy Citation Text

Jinhe Fan, Jing Wu, Maolin He. Super-Resolution Computed Tomography Reconstruction of Residual Attention Aggregation Dual Regression Network[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0210009

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites