Efficient and accurate recognition of railway infrastructure objects is essential for ensuring safe and reliable railway operations. This paper proposes the utilization of the YOLOv8 deep learning model for the instance segmentation of railway infrastructure objects. The YOLOv8 model employs a powerful deep neural network architecture with multiple convolutional layers, enabling real-time object detection and segmentation in a single pass. To train the YOLOv8 model, a comprehensive dataset consisting of 20,000 high-resolution images was created. Manual annotation was performed using polygonal segments to precisely capture the shapes and boundaries of the railway infrastructure objects. The dataset encompasses 40 classes of objects commonly encountered in railway environments, including railway tracks, switch components, kilometer posts, boundary posts, traffic lights, and various other objects. Evaluation of the YOLOv8 model demonstrated its effectiveness, achieving a mean Average Precision (mAP) of 0.8 at an intersection-over-union (IoU) threshold of 0.5 and an mAP of 0.54 from an IoU threshold of 0.5 to 0.95. Moreover, the proposed model demonstrated exceptional computational efficiency, achieving processing time of a mere 33 ms per frame when deployed on the NVIDIA RTX A5000 graphics accelerator. The proposed approach addresses the limitations of classical computer vision methods, providing robust and real-time segmentation of railway infrastructure objects. The findings contribute to the development of more reliable and efficient railway infrastructure recognition systems, enhancing safety and operational performance in railway environments. The fast-processing time of a single frame enables the method's integration into real-time systems, further extending its practical applicability.