摘要
霾监测是环境治理中的关键技术之一。目前地面观测站进行霾监测的耗费较大,基于多光谱遥感的霾识别精度较低。将深度学习用于高光谱遥感数据的霾监测,提出一种基于深度残差网络的高光谱霾监测方法,利用深度网络提取霾光谱曲线特征,再使用残差学习等方法降低网络训练难度,得到了霾监测模型。苏州地区Hyperion高光谱数据集上的实验表明,与其他遥感霾监测方法相比,所提方法的霾识别精度更高。
Haze monitoring is one of the key technologies for environmental governance. At present, the cost of the ground haze monitoring is very high and the accuracy of the multispectral remote sensing haze monitoring is low. The hyperspectral sensing data haze monitoring is studied by deep learning. A hyperspectral haze monitoring algorithm based on deep residual network is presented. The features of haze hyperspectral curves are obtained with the deep network. The difficulty of the network training is decreased with the residual leaning method, and a haze monitoring model is achieved. The experimental results of the Suzhou Hyperion hyperspectral data sets show that, compared with other methods of remote haze monitoring, the proposed method has higher recognition accuracy in haze monitoring.
出处
《光学学报》
EI
CAS
CSCD
北大核心
2017年第11期306-316,共11页
Acta Optica Sinica
基金
国家自然科学基金(U1406404)
上海市军民结合项目(沪经信军[2014年]495号)
关键词
遥感
大气污染监测
霾监测
深度残差网络
高光谱遥感
深度学习
机器学习
remote sensing
air pollution monitoring
haze monitoring
deep residual network
hyperspectral remote sensing
deep learning
machine learning
