Abstract: To address the problem that the signal-to-noise ratio of cryo-electron microscopy images is extremely low and the existing denoising methods cannot effectively remove complex noise, a cryo-electron microscopy image denoising method based on improved U-Net is proposed. The improved method used FCN to build a noise mapping module and embedded multi-scale concatenation and wide-activation dense residual blocks in the original U-Net network, which not only improved the generalization ability of the network but also enabled the model to better extract and recover feature information to achieve high-quality cryo-electron microscopy image denoising. The introduction of the full-variance loss function was used to protect the particle detail information in the output image. Experimental results show that the method can effectively remove background noise while better recovering particle details, the signal-to-noise ratio (SNR) is also optimal, and the number of picking particle positives is improved compared with the comparison method.