IFRG Diffusion: A Diverse Railroad Image Generation Based on Diffusion Models*

The advancement of intelligent rail transportation systems is significantly hindered by the scarcity of large-scale, high-quality datasets. To address this challenge, we propose a novel framework for generating rail transportation images based on fine-tuned diffusion models. Our framework aims to produce high-quality and diverse rail transportation data from a limited set of image inputs. Specifically, we introduce a dual-branch feature extraction network that effectively fuses low-level image features with high-level semantic information, thus enhancing feature utilization under constrained data conditions. We further incorporate a compression encoding-decoding module to facilitate information exchange between the input data and the diffusion model, thereby improving the detail and texture quality of generated images. By leveraging group supervision, the proposed method mitigates overfitting on anomalous data, enabling swift and accurate learning of new categories from minimal data and generating diverse, rich rail images. Experimental results demonstrate that our framework not only produces more realistic and logically consistent rail transportation images but also exhibits robust generalization and adaptability.