BMAD-Net: An attention mechanism network using block match for laboratory X-ray Microscopy denoising

CNNs have been widely used for image denoising. An attention mechanism network using block match for laboratory X-ray Microscopy denoising is proposed. The similar block is introduced into a CNN model to capture the global information connections of the images, which complements the information loss by weakening long-distance dependencies with increased convolution network depth. A multi-scale convolutional subnetwork is designed to enhance the receptive field as the noise feature extraction module. Then, by reassigning weights to the feature channels, a residual channel attention block is designed to improve the feature channel’s availability. Both experiments on bronze coins and fossils show that the proposed method can perform effective image denoising within 0.1s exposure time, and the image quality is comparable with those within 3s exposure time, reducing the existing acquisition time. •Propose a network that utilizes block matching and apply it for the first time to X-ray microscopy denoising.•Utilize the self-similarity of images to capture the global information connections.•Multiscale convolutional subnetwork is designed to enhance the receptive field as the feature extraction module.•Residual channel attention block is designed to improve the feature channel’s availability.•This method improves the image quality of 0.1 s exposure time to comparable to 3 s.