Extending Channel Attention to Kernel Space for Fluorescence Microscopy Imaging Denoising

Deep learning has demonstrated significant potential in fluorescence microscopy imaging. However, most existing methods primarily focus on enhancing channel or spatial features, overlooking the capabilities of kernel space. This limitation restricts the network's ability to recover fine structural and details, especially in noisy or degraded imaging conditions. In this work, we propose iKUNet-RCAN, a novel architecture that integrates kernel and channel attention mechanisms. By explicitly capturing kernel space dependencies, our model enhances feature representation with minimal computational cost. Experimental results across multiple microscopy modes (confocal, widefield, two-photon) demonstrate superior performance in image quality and reconstruction robustness compared to existing methods.