This article proposes a single soldier behavior analysis model based on neural networks and data-driven approaches. Mainly based on the following considerations: firstly, constructing key target (individual soldier) models through open source intelligence such as network videos, unmanned aerial vehicle reconnaissance videos, and fixed-point monitoring videos to solve the current problem of lack of target modeling methods. The second is to assist commanders in predicting the behavior of key targets by constructing their movement trajectories. The focus is on studying the following three aspects: firstly, building a key target model in real-time scenarios based on the IMAP ER algorithm; Secondly, by optimizing the residual neural network calculation method, key target model behavior descriptors can be extracted for real-time autonomous training and learning; Thirdly, based on consistency correspondence, establish the movement trajectory of key targets from different perspectives.