Enhancement of Biomass Material Characterization Images Using an Improved U-Net

• DOI: 10.32604/cmc.2022.024779
• Publication Date: 2022-02-24T07:35:42Z
• Authors: Zuozheng Lian; Hong Zhao; Qianjun Zhang; Haizhen Wang; E. Erdun
• Keywords: Deblurring; Convolution (computer science)

For scanning electron microscopes with high resolution and a strong electric field, biomass materials under observation are prone to radiation damage from the beam. This results in blurred or non-viable images, which affect further of material microscopic morphology characterization. Restoring images their original sharpness is still challenging problem image processing. Traditional methods can't effectively separate context dependency texture information, effect enhancement deblurring, gradient disappearance during model training, resulting great difficulty training. In this paper, we propose use an improved U-Net (U-shaped Convolutional Neural Network) achieve for characterization restore sharpness. The main work as follows: depthwise separable convolution instead standard reduce computation effort parameters; embedding wavelet transform into structure thereby improving reconstruction quality; using dense multi-receptive field channel modules extract detail better transmitting features network gradients, experiments show that proposed paper suitable effective enhanced deblurring images. PSNR (Peak Signal-to-noise Ratio) SSIM (Structural Similarity) well.