In this paper, aiming at the problems of low accuracy and high false detection rate of railway foreign object intrusion detection in complex environment, a railway foreign object intrusion algorithm MS YOLOv7-Tiny algorithm based on improved YOLOv7-Tiny is proposed. Firstly, MLCA lightweight attention mechanism is introduced to enhance the ability of capturing and locating the features of encroachment targets. Secondly, by adding a small target detection layer, the feature information of different scales is fully integrated to capture the detailed features of small targets, and the multi-scale target detection capability of the network is enhanced. Finally, through relevant experiments on self-built data sets, compared with the original YOLOv7-Tiny algorithm, the detection accuracy and recall rate of the proposed method are improved by 5.2%, 4.5% and 1.4%, respectively, with higher detection performance and a wider range of applicable scenarios.