基于下采样的特征融合遥感图像语义分割

本文提出了一种基于下采样的特征融合遥感图像语义分割模型,该模型在编解码结构基础上,将高分辨率原始图像引入“下采样”模块提取低级语义特征,在此基础上,将输出的低级语义特征通过MobileNetV2和空间金字塔池化进一步提取多尺度高级语义细节特征,然后,将这些高级语义特征和直接从下采样模块提取的低级语义特征融合并进行特征图分割.最后,在“CCF卫星影像的AI分类与识别竞赛”的数据集上取得了93%的训练准确率以及91%的预测准确率.

Feature Fusion Based Semantic Segmentation of Remote Sensing Image Using Down-Sampling

This paper proposes a model of feature fusion based semantic segmentation of remote sensing image.Based on the coding and decoding structure,high-resolution original images were firstly introduced into“down-sampling”module in order to get low-level semantic features,from which,detailed features of multiscale high-level semantic are then obtained by MobileNetV2 and atrous spatial pyramid pooling(ASPP).Moreover,the high-level semantic features and the low-level semantic features directly extracted from the down-sampling module were fused and then segmented of characteristic pattern.Finally,by the CCF Contest for AI Classification and Identification of Satellite Image,a training accuracy of 93%and a forecast accuracy of 91%were obtained.