Learning the perception, cognition, and complex behavior mechanisms of neural networks through dynamic observation of large volume neurons in the brain is a fundamental issue in neuroscience. Light field microscopy, which allows for three-dimensional imaging at video frame rate, has become a technological choice for real-time observation of neural activity. An overview of the current imaging methods of light field microscopy is provided in this article, with a focus on new algorithms.Compared with traditional iterative reconstruction methods, using light field microscopy to image neuronal activity requires the development of new deep learning based algorithms that fully utilize the prior embedded in physical and optical models, integrate model driven and data driven algorithms, enhance the interpretability and generalization ability of algorithms, and achieve fast and robust light field microscopy imaging.