基于多层次感知网络的GF-2遥感影像建筑物提取

Buildings extraction of GF-2 remote sensing image based on multi-layer perception network

高分辨率遥感影像建筑物提取任务在城市规划、城镇化进程等领域发挥着重要作用。针对现有的深度学习提取方法存在浅层特征未得到有效利用、小目标信息容易丢失等问题,提出了一种多层次感知网络。该网络利用密集连接机制充分提取特征信息,并构建平行结构保留不同特征分辨率的空间信息,增强不同深度、尺度特征信息,减少细节特征的丢失;同时利用空洞空间金字塔模块获取不同感受野信息,提取不同尺度下的深层建筑特征。实验结果表明,该方法在GF-2遥感影像建筑物提取中,总体精度为97.19%、交并比为74.33%、综合评价指标为85.43%,各指标均高于传统方法与其他深度学习方法;此外,应对多源遥感影像的建筑物仍具有良好的提取效果,体现了本文方法的实用性。

The task of extracting buildings with high-resolution remote sensing image plays an important role in urban planning and urbanization.In view of the problems of existing deep learning extraction methods,for example,the shallow features can’t been used effectively and small target information is easily lost,this paper proposes a multi-level perceptual network.This network uses dense connection mechanism to fully extract feature information,and constructs parallel structure to retain spatial information of different feature resolution and enhance feature information of different depth and scale in order to reduce the loss of detail feature.At the same time,the ASPP module is used to obtain the information of different receptive fields and extract the deep architectural features at different scales.The experimental results show that the overall accuracy of the proposed method is 97.19%,intersection over union is 74.33%and the F1 score is 85.43%in the buildings extraction of GF-2 remote sensing image,all of which are higher than those of the traditional method and other deep learning methods.In addition,buildings with multi-source remote sensing images still have good extraction effect,which reflects the practicability of the method presented in this paper.