基于CNN模型的遥感影像液气贮存设备识别  

Research on Image Recognition of Liquid Gas Storage Equipment Based on CNN Model

作　　者：李梦梦 陈超[1] 陶旸[1] LI Mengmeng;CHEN Chao;TAO Yang(Provincial Geomatics Centre of Jiangsu,Nanjing 210013,China)

机构地区：[1]江苏省基础地理信息中心,江苏南京210013

出　　处：《地理空间信息》2023年第5期110-113,共4页Geospatial Information

基　　金：江苏省测绘地理信息科研资助项目(JSCHKY201924)。

摘　　要：基于卷积神经网络(CNN)进行遥感影像液气贮存设备提取。基于0.3 m、0.5 m、2.1 m分辨率的多源遥感影像,通过训练CNN模型来提取液气贮存设备,检测精度F值分别为0.78、0.77和0.36;基于0.3 m、0.5 m、1 m分辨率的无人机影像,训练得到不同的CNN模型,迁移学习至0.8 m分辨率的高分影像上进行液气贮存设备识别,检测精度F值分别为0.08、0.73和0.71。结果表明,CNN模型可有效提取液气贮存设备,且训练样本包含的信息越均一,识别精度越高;CNN模型对多源遥感影像具有较好的迁移学习能力,在遥感影像典型目标识别与提取方面具有较好的应用前景。In this paper,we extracted the liquid gas storage equipment of remote sensing images based on convolutional neural network(CNN).For multi-source remote sensing images with three different resolutions of 0.3 m,0.5 m and 2.1 m,we trained CNN model to extract liquid gas storage equipment.The detection accuracy F values were 0.78,0.77 and 0.36,respectively.Then,we obtained different CNN models based on the training of UAV images with 0.3 m,0.5 m and 1 m resolution,and transferred them to the 0.8 m resolution high-resolution image for identification of liquid gas storage equipment.The detection accuracy F values were 0.08,0.73 and 0.71,respectively.The experimental results show that CNN model can effectively extract liquid gas storage equipment,and the more uniform the information contained in the training samples,the higher the recognition accuracy.For multi-source remote sensing images,CNN model has better transfer learning capabilities,and a good application prospect in the recognition and extraction of typical targets in remote sensing images.

关 键 词：CNN 迁移学习 遥感目标检测 液气贮存设备 

分 类 号：P237[天文地球—摄影测量与遥感]