I. Introduction
Deep learning technology helps promote the development of state-of-the-art techniques in the field of railway foreign object intrusion detection, and has become a useful tool for pioneering research[1]. This Convolutional neural network (CNN) algorithms have enabled the application of deep learning technology to railway foreign object intrusion detection, resulting in high detection accuracy and relatively low false alarm rates[2]. In the field of railway foreign object intrusion detection, convolutional neural network methods eliminate the cumbersome manual feature extraction process compared with traditional algorithms and can achieve better classification results. However, the training and fitting of convolutional neural network models often rely on large datasets, sufficient computing resources, and efficient neural network structures. Data, hardware computing power, and algorithms all affect the task processing effect of convolutional neural network models, making efficient convolutional neural network design and implementation particularly important [3].