Aiming at the problem that most of the existing smoke detection methods are only suitable for the environment with sufficient light and the detection effect is poor in the low light environment, this paper proposes a low light smoke detection method based on FSSD. Firstly, the video frames containing moving objects are extracted based on single gaussian modeling method; secondly, the low contrast and low signal-to-noise ratio characteristics of low light level smoke image will cause difficulties in target detection, so the image preprocessing method combining the contrast limited adaptive histgram equalization and median filter is designed. Finally, in order to strengthen the early warning ability, the FSSD network that is conducive to detecting small targets is adopted, convolutional block attention module is embedded in the output of the network, which can strengthen the important feature information and improves the accuracy of target detection. On the low light smoke dataset, the Recall, Precision, and F1 of the proposed method reached 97.5%, 93.3% and 95.4%, indicating that the method is effective and can be applied to smoke detection in low light environments.