A Remote Sensing Image Segmentation Network Based on Multimodal Feature Fusion Enhancement

The development of high-resolution remote sensing image technology plays a promoting role in many fields such as urban land planning, agricultural production and environmental monitoring. However, most traditional remote sensing image semantic segmentation networks only rely on single-modal data, while high-resolution remote sensing images usually have complex background information and cover massive amounts of data. This paper proposes a segmentation network MCTUNet based on multimodal feature fusion enhancement, which is constructed by combining convolutional neural networks (CNN) and visual converters (VIT) to improve the extraction of local detail features and global semantic modeling. The MFEF module is innovatively proposed to enhance the respective representation capabilities of multimodal features and the interaction capabilities among different modal features, aiming to utilize the complementary advantages of visible light images and DSM images more effectively. The decoding layer adopts transposed convolution to achieve upsampling to avoid the loss of feature information. The experimental results show that on the high-resolution remote sensing dataset Vaihingen, the performance of MCTUNet is superior to that of other multimodal fusion semantic segmentation algorithms.