BPSNet: Edge-preserving SAR image denoiser using attention based bit plane slicing network

Synthetic aperture radar (SAR) captures object reflectivity through electromagnetic signals, but random phase instabilities introduce multiplicative speckle noise that deteriorates image quality and complicates visual analysis. Despite extensive research, balancing speckle suppression with preservation of fine details remains challenging. To address this problem, a bit-plane slicing - based Convolutional Neural Network(CNN) is proposed in this paper, that first decomposes SAR images into Most Significant Bits (MSB) and Least Significant Bits (LSB) and then a Sobel edge operator is applied to the MSB to highlight structural details, while a Pixel Attention Block (PAB) is applied to LSB to adaptively emphasizes informative pixels to enhance detail preservation. This integration of bit-level decomposition, edge enhancement, and pixel-level attention enhances despeckling performance, producing images with improved visual quality and sharper edges, as validated by experimental results.