降噪
人工智能
信噪比(成像)
模式识别(心理学)
算法
噪音(视频)
计算机科学
峰值信噪比
均方误差
图像(数学)
数学
统计
电信
作者
Quan Tang,Jiaqi Hu,Jinna Chen,Chenlong Xue,Junfan Chen,Hong Dang,Dan Lu,Huanhuan Liu,Qizhen Sun,Qihua Xiong,Longqing Cong,Perry Ping Shum
标识
DOI:10.1109/acp55869.2022.10088904
摘要
For the problem of low Signal-to-Noise Ratio (SNR) of the image reconstructed from Raman spectra, this paper proposes a two-stage denoising algorithm based on deep learning, including spectrum denoising and image denoising. Because spectra and images of the same sample are scarce, the spectrum denoising algorithm and the image denoising one are trained on two irrelevant dataset. Denoising and baseline correction are performed on the raw Raman spectra using 1D-ResUNet. The low SNR images are denoised by Reversed Convolutional Block Attention Module UNet (RCBAM-UNet). Experimental results showed that the Mean Squared Error (MSE) between the denoised spectrum processed by our method and the ground truth decreased by 10 3 -10 4 , when compared to the raw data. The quantitative results of image denoising on Structural Similarity (SSIM) and Peak Signal-to-Noise Ratio (PSNR) are 98.4% and 39.7, respectively. The performance of our two-stage algorithm on the testing dataset (an independent dataset) showed increased SNR and Contrast to Noise Ratio (CNR) in Raman imaging denoising. Our study provides an efficient method for the improvement of Raman image quality.
科研通智能强力驱动
Strongly Powered by AbleSci AI