Historical documents and palm leaves contain precious knowledge but are fragile and degrading, in a broken state. To preserve these invaluable resources, digitization is essential, with line segmentation being a crucial subtask in this process, this addresses the text format and orientation of the palm leaf document, maintaining the quality of the text lines. In our work, palm leaf document text lines are segmented using Segformer, a segmentation framework based on the multi-head attention mechanism of Vision Transformer (ViT). The key benefit ensures in handling complex layouts, dealing with degraded text and manages variations in text flow, these decrease the line orientation complexity and increase OCR accuracy. The lines are pre-defined for training images through a line annotation tool, which allows unwanted noise or textual content to be ignored and grasped in detail. The experimental analysis is based on Intersection over Union (IU) metrics in comparison with Fully Convolutional Network (FCN) benefiting Segformer. The Segformer trained on 200 images has given a higher IU score than Segformer trained on 50 images, showcasing increase in the dataset the more the model. The experimental results show that the Segformer gives the best results with an average IU score of 74.27%.