Integrating local and global information for image deblurring in flotation froth scenes

Abstract

Identifying and deblurring froth images in flotation conditions is of significant importance. Due to the rapid displacement of froth, images collected on-site in flotation processes are often motion-blurred. To address this issue, we propose a multi-scale nonlinear activation free block to enhance the extraction of spatial details and contextual information in the encoder. We also remove unnecessary nonlinear activation functions to simplify system complexity. Next, we design a convolution and transformer hybrid block in the decoder to extract and integrate different local and global blur information for accurate recovery. Compared with existing advanced image deblurring networks, experimental results show that our network has significant advantages in deblurring images in flotation froth scenarios. Specifically, in terms of PSNR and SSIM indicators, the deblurring accuracy of this network reaches 33.6138 dB and 0.9483, respectively.