Optimization method for TDI-CCD image noise suppression based on improved transformer algorithm

In low earth orbit optical remote sensing satellite detection missions, the multi-chip splicing linear array time delayed integration charge coupled device (TDI-CCD) technology has been widely adopted. However, this technology faces the challenge of crosstalk noise. In response to this issue, this study proposes an innovative TDI-CCD image noise suppression scheme based on an improved Transformer algorithm. Firstly, the network architecture incorporates Transformer mechanism at the encoder decoder level, aiming to solve the long-distance dependency problem in large-scale image data and improve the efficiency of denoising processing. Secondly, to address the issue of Transformer neglecting spatial relationships between pixels and causing local detail loss, we designed a Feature Refinement Block (FRB) during the feature reconstruction stage. This module adopts a serial structure and applies nonlinear transformations layer by layer to enhance the recognition ability of local features in noisy and complex images. At the same time, a multi-scale attention block (MAB) is constructed, which adopts a parallel dual path design to jointly model spatial attention and channel attention, effectively capturing and weighting image features of different scales, thereby improving the model’s recognition ability for multi-scale features. Through a series of simulation experiments, the algorithm proposed in this study demonstrates significant performance advantages in balancing global information and local details.