Noise and false exposure are important reasons for image information loss. Aiming at the problem that it is difficult to control exposure and noise in low-light image enhancement, a multi-branch convolutional network is designed and the loss function is modified to enhance image details and remove noise. Inspired by the MSR algorithm, the algorithm designs a three-branch five-layer convolution kernel network structure, and avoids underexposure or overexposure that may occur in a single branch by means of multi-branch averaging. In order to balance noise removal and detail preservation, the SSIM loss function is introduced, combined with the U-Net like network structure of the residual network, to compress image features to reduce the proportion of noise while retaining complete structural information. Experimental results verify the effectiveness of the proposed network.