This study mainly deals with the improvement of the image enhancement network based on zero-reference deep curve estimation (Zero-DCE). Upon the image convolution in each layer, the image will lose some detailed content and confront noise problems, and thus an improved network structure is proposed. The convolutional layer retains the main content of the image, and the deconvolutional layer is added to compensate for the detail loss. The feature map of the convolutional layer is transmitted to the deconvolutional layer, which can help the decoder to obtain more image details for a cleaner image. In addition, a residual network is introduced to make a difference between the input noise image and the output clean image to learn a residual error, improving image clarity and reducing noise. Finally, the image quality evaluation methods, i.e., the peak signal to noise ratio (PSNR) and the structural similarity index measure (SSIM), and the Fourier transform are used for testing and analysis. The results show that the improved structure proposed can increase the image details and achieve the effect of noise reduction.