一种面向典型农村道路遥感提取的深度学习方法

• Format: Patent
• Language: Chinese
• Published: 18.10.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

本发明公开了一种面向典型农村道路遥感提取的深度学习方法，包括：(1)获取用于典型农村道路提取的高分辨率遥感影像，并进行预处理，包括辐射定标、大气校正、几何校正、波段融合；(2)依据预处理后的高分遥感影像进行人工目视解译，获取道路矢量数据，并制作模型的训练与测试数据集；(3)在U-Net模型中加入优化残差模块、全局上下文注意力机制模块和DUpsampling模块，提出GDU-Net模型；(4)利用训练数据集进行模型训练，然后将测试数据集输入到模型中，进行道路提取与结果评价。本发明不仅能够正确提取农村道路的边界，而且增强了提取结果的完整性，显著提升了典型农村道路的提取精度，具有较好的应用价值。 The invention discloses a deep learning method for typical rural road remote sensing extraction, and the method comprises the steps: (1), obtaining a high-resolution remote sensing image for typical rural road extraction, and carrying out the preprocessing, including radiometric calibration, atmospheric correction, geometric correction and wave band fusion; (2) performing artificial visual interpretation according to the preprocessed high-resolution remote sensing image to obtain road vector data, and making a training and testing data set of the model; (3) adding an optimization residual module, a global context attention mechanism module and a DUpsampling module into the U-Net model, and proposing a GDU-Net model;