基于改进Deeplabv3+网络的遥感图像选站选线语义分割

Semantic Segmentation of Station Selection and Line Selection in Remote Sensing Image Based on Improved Deeplabv3+Network

在输电线选站选线设计中,所选范围内的地物信息至关重要,为使计算机能够自动识别并标记遥感图像中的地物信息,辅助人工选线,开展设计智能选站选线系统。选择Deeplabv3+网络模型分割遥感图像中的关键性地物,并在此基础上做出改进。在Deeplabv3+的解码区网络中,充分利用主干网络产生的多尺度特征信息,并对深层特征进行更细化的上采样操作。基于AID(aerial image data)数据集进行地物分割实验对比,实验表明,改进后的Deeplabv3+可以更精细地分割出地物类别与范围,为智能化选站选线设计提供更有力的支持。

In the design of station selection and line selection of power transmission line,the ground object information within the selected range is very important.In order to enable the computer to automatically recognize and mark the ground object information in the remote sensing image and assist the manual selection of lines,the intelligent system of line selection and station selection is designed.Deeplabv3+network model is chosen to obtain the key features of remote sensing image segmentation and the improvements network model is also investigated.In the decoding region network of Deeplabv3+,the multi-scale feature information generated by the backbone network is fully utilized,and the up-sampling operation of the deep feature is more detailed.Based on aerial image data(AID)set,ground object segmentation experiments are conducted.The experiments show that the improved Deeplabv3+can more accurately segment the category and range of ground object,which provides powerful support for the design of intelligent station selection and line selection.