稳定扩散模型在遥感图像生成中的应用  

The application of stable diffusion in the generation of remote sensing imagery

作　　者：周登继 刘国阳 黄济富 蒙静远 Zhou Dengji;Liu Guoyang;Huang Jifu;Meng Jingyuan(Power China Huadong Engineering Corporation Limited,Hangzhou,Zhejiang 311122,China;Zhejiang Huadong Engineering Digital Technology)

机构地区：[1]中国电建集团华东勘测设计研究院有限公司,浙江杭州311122 [2]浙江华东工程数字技术有限公司

出　　处：《计算机时代》2025年第1期11-15,共5页Computer Era

摘　　要：深度学习中遥感影像解译需要大量高质量遥感数据,但获取大量高分辨率的遥感数据成本高昂。为高效、经济地解决模型训练数据匮乏的问题,本文利用稳定扩散模型(Stable Diffusion)实现了遥感图像的生成。该方法利用DreamBooth技术和控制网络(ControlNet),通过文字描述从噪声端到端生成多传感器遥感影像。实验结果表明,该方法不仅能够有效提高训练数据的多样性,而且生成的数据质量接近真实数据水平,解决了训练数据不足的问题。Remote sensing image interpretation in deep learning requires a large amount of high-quality remote sensing data,but it is costly to obtain a large amount of high-resolution remote sensing data.To solve the problem of lack of model training data efficiently and economically,this paper uses the stable diffusion model(SD)to generate remote sensing images.This method utilizes DreamBooth and ControlNet to generate multi-sensor remote sensing images from noise end to end through text description.The experimental results show that this method not only can effectively improve the diversity of training data,but also the quality of the generated data is close to the level of real data,which solves the problem of insufficient training data.

关 键 词：稳定扩散模型 DreamBooth CONTROLNET 深度学习 遥感图像生成 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]