Improving Road Extraction in Hyperspectral Data with Deep Learning Models

Accurately extracting road networks from hyperspectral data using convolutional neural network models is challenging due to various factors such as occlusion, changing lighting conditions, and blur. To address this issue, this paper proposes a new model that combines the advantages of U-net and Transformer architectures. This hybrid model effectively captures both local and long-range features, thus improving the accuracy and efficiency of road extraction. Evaluation of the method is performed on the AeroRIT hyperspectral dataset using performance metrics such as overall accuracy, average per-class accuracy, and average Jaccard index. Compared with traditional convolutional neural network models such as U-net. The results show that the proposed method improves the average per-class accuracy by more than 18% over the traditional methods, demonstrating its potential to optimize road extraction from hyperspectral data. Further research can focus on improving the accuracy and efficiency of road network extraction from hyperspectral data.