Multimode fibers (MMFs) have great potential for endoscopic imaging due to the high number of modes and small core diameter. Deep learning based on neural networks has received increasing attention in the field of scattering image reconstruction. However, most of the research lies in designing complex network architectures to improve reconstruction, and these network models are not capable of reconstructing images in low ambient light. In this paper, a lightweight generative adversarial network (GAN) model combined with a histogram specification algorithm is designed to reconstruct dark-field speckles through MMF. Experimental results show that the reconstruction results of our algorithm have better metrics. Moreover, the model exhibits excellent cross-domain generalization ability with regards to the Fashion-MNIST dataset. It is worth mentioning that we probe and find that the speckles after inactivation still have the ability to be reconstructed, which increases the robustness of the model.