Semantic segmentation of remotely sensed images is of great significance in the fields of geographic information systems (GIS) and environmental monitoring. In this paper, a semantic segmentation method for remote sensing images based on U-Net network is proposed, aiming to achieve efficient and accurate pixel-level classification. Firstly, data enhancement is performed on remote sensing images and elastic deformation technique is used to increase the diversity of training data. Secondly, the overlap-tile strategy is utilised for image cutting to provide contextual information to the edges of the segmented region and improve the prediction accuracy. Finally, the U-Net network model is designed and implemented and trained using the weighted loss function with cross-entropy loss and Dice loss function. The experimental results show that the method achieves high segmentation accuracy on several remote sensing image datasets, with a PA value of 94.1%, an average CPA value of 89.8%, and an average IOU value of 86.1%. In addition, a GUI interface is developed in this paper to facilitate the user’s image segmentation operation and model training visualisation. This study provides new ideas and methods for semantic segmentation of remote sensing images, which has important theoretical significance and application value.