TICNet: Three-Branch Real-Time Semantic Segmentation Network with Intensive Compensation of Railway Track Scenes

With the rapid development of railway traffic system, real-time semantic segmentation plays a crucial role in railway track scene monitoring. However, most of the existing methods use lightweight convolution to reduce the computational complexity, which results in reduced accuracy. To solve this problem, this paper proposes a real-time semantic segmentation network called TICNet, which is specifically used for accurate segmentation of railway track scenes. TICNet uses a novel three-branch structure designed for efficient feature extraction and accurate segmentation. Results on the RailSem19 dataset show that TICNet has a segmentation accuracy of 62.36% mIoU with an inference speed of 81.2 FPS, which is significantly better than many existing methods.