For the task of HDR (High Dynamic Range) image generation, in order to solve the problems of long acquisition time of multi-exposure images, inter-frame offset in dynamic scenes, and large number of algorithm parameters and computation of existing methods, this paper proposes a lightweight HDR image fusion algorithm based on the linear logarithmic response camera, and acquires a multi-gain grayscale image dataset. Firstly, the improved multi-scale residual module is used to extract the multilevel features of the input image and enhance the feature dimension. Secondly, the multilevel feature input is introduced into the Attention-Unet structure with depth-separable convolution to extract the multilevel information in the features and fuse the features. Thirdly, the point-by-point convolution is used to fuse the depth features of the image, and to output high dynamic range images compatible with standard display devices without additional tone mapping. Finally, the performance of each ablation structure is compared with the number of parameters and computation, and the optimal solution that guarantees the fusion effect while keeping the network lightweight is obtained. The experimental results show that the algorithm proposed in this paper has better performance in both subjective visual effect and objective evaluation index, with MEF-SSIM of 0.986 6, VIF of 1.76, AG of 3.94, and SF of 14.32.The high dynamic image fusion algorithm proposed in this paper maintains the excellent fusion effect and robustness in the case of the significant difference between multi-gain images, and has the lightweight, the number of model parameters is only 0.612M, and the computational complexity is 7.254 GFLOPs.