一种深度卷积神经网络土地利用场景照片的分类方法

A method of deep convolutional neural network photo classification for landuse scenes

国土调查多角度实景举证照片具有视角多、分辨率高、层次丰富和剖面清晰的特点,透视且细致地刻画了土地利用图斑赋存状况和场景,弥补了遥感影像单一天顶视角的不足。本文基于语义分割提出了一种深度卷积神经网络(DCNN)实景照片土地利用场景分类方法,多语义标记照片场景,语义组合智能判定照片土地利用类别。该方法成功地应用在第三次国土调查照片自动核查工作中,减轻了人工判读工作量,提高了土地利用场景自动识别的精度。

The multi-angle real-life proofed photos of the national land survey have multiple characteristics:high resolution,rich layers,and clear sections.The photos are accurate and detailed depictions of the occurrence status and scenes of landuse maps,which makes up for the lack of ground perspective of remote sensing images.Based on semantic segmentation,this paper proposes an innovative landuse scene classification method with a deep convolutional neural network(DCNN)for ground perspective photos.The method annotates photo scenes with multi-semantic technique and determines categories of photos of landuse with semantic composition.Successfully applied in the third automatic survey of land survey photos,this method reduces the workload of manual interpretation and improved the accuracy of automatic identification of landuse scenes.