In response to the problem that military-related rumor detection is neglected in online rumour detection, which affects the online ecology and also brings challenges to the image of the army and the psychology of officers and soldiers, we analyze the characteristics of military-related rumours and their linguistic features, and try to apply Bi-directional long short term memory (Bi-LSTM) and attention mechanism, We propose a military-related rumour detection model based on Bi-LSTM + Attention mechanism. The model is based on the model comparison of public datasets, and the military-related datasets established to complete the experimental tests, and the results found that its accuracy rate reached 97.8%, which has a better detection effect.