Distributed fiber-optic sensing technology can detect system parameters such as temperature, strain, and vibration by demodulating the change in scattered light in optical fiber to achieve distributed measurement. The weak scattered-light signal results in low signal-to-noise ratios for a system and limited sensing performance. In recent years, researchers have focused on enhancing the performance of distributed fiber-optic sensing systems using machine learning. Herein, from the perspectives of data extraction, noise removal, and resolution enhancement, the progress of machine-learning technology in the research and development of distributed fiber-optic sensing in the past decade is described comprehensively, and an analysis and comparison with technology outlook is performed. Different types of machine-learning techniques offer different degrees of improvement to the processing speed. Among them, denoising neural networks are more effective in improving the signal-to-noise ratio, as shown by their superior noise suppression and ability to further improve the signal quality. Additionally, convolutional neural networks are outstanding in terms of improving spatial resolutions and can capture and resolve complex spatial features more accurately. These results are expected to further promote the subsequent in-depth application of machine-learning techniques to various distributed fiber-optic sensing technologies.