error correctionIt can be seen that our method has achieved better results than other technologies Table Ⅱ shows the test results of the images without turnouts after the training of each model We can

It can be observed that our method has achieved superior results compared to other technologies. Table II presents the test results of images without turnouts after training each model. We can notice that, within the same model, the test accuracy for rail is the highest while that for fasteners is the lowest. This is because the rail has a simple and easily identifiable shape, whereas the fasteners have a complex shape and varying numbers, making them difficult to identify. Among the same category, the RTMOSeg model demonstrates the highest accuracy in the testing of fasteners and sleepers, while TransUnet performs the best in rail testing. However, the difference in accuracy between RTMOSeg and TransUnet is minimal, with less than a 0.3% variation. These results indicate that the proposed RTMOSeg model provides better segmentation compared to other models. Additionally, in the test results of RTMOSeg, there is little difference in the accuracy of fasteners, sleepers, and rails, further demonstrating that RTMOSeg is effective for both simple and complex targets. Table III displays the test results of images including turnouts after training each model. We observe that, within the same model, the accuracy of rail remains the highest, but the accuracy of fasteners is higher than that of sleepers. This is because the shape of the sleeper becomes more complex in images with turnouts. Within the same category, the RTMOSeg model achieves the highest accuracy in the testing of fasteners, sleepers, and rails, indicating superior performance. Moreover, the test accuracy of RTMOSeg on dataset #1 and dataset #2 is similar, further demonstrating the strong robustness of RTMOSeg