摘要
建筑物边缘检测是提取建筑物信息最直接有效的方法,近几年,卷积神经网络被广泛应用于建筑物边缘检测研究,其中RCF网络被证明是应用于建筑物边缘检测的效果较好的卷积神经网络。然而,RCF网络在建筑物边缘检测的过程中,上采样过程采用一步双线性插值算法,上采样结构过于简单,导致产生了在深层网络特征层小尺寸图像特征直接上采样至大尺寸图像的条件下误差过大的问题。文章提出了阶梯式上采样结构以改进RCF网络,该方法能够有效减少一次双线性插值算法带来的误差,实验证明该方法能够有效提高RCF网络在建筑物边缘检测上的结果精度,显著增加输出结果图像的清晰度。
Building edge detection is the most direct and effective method to extract building information,in recent years,convolutional neural networks have been widely used in building edge detection research,among which RCF network has been proved to be a convolutional neural network with good effect applied to building edge detection.However,in the process of RCF network detection at the edge of the building,there is a problem that the upsampling process adopts a one-step bilinear interpolation algorithm,and the upsampling structure is too simple,resulting in too large error under the condition that the features of the small-size image of the deep network feature layer are directly upsampled to the large-size image.In this paper,a stepped upsampling structure is proposed to improve the RCF network,which can effectively reduce the error caused by the bilinear interpolation algorithm,and the experimental results show that the method can effectively improve the accuracy of the RCF network in the detection of building edges,and significantly increase the clarity of the output images.
作者
刘佳蕙
苏杭
Liu Jiahui;Su Hang(Jiangsu Normal University,Xuzhou 221116,China)
出处
《无线互联科技》
2023年第6期80-84,共5页
Wireless Internet Technology
基金
江苏省研究生科研与实践创新计划项目,项目名称:电力设施环境AI遥感监测研究,项目编号:KYCX21_2627。
关键词
卷积神经网络
RCF网络
建筑物边缘检测
阶梯上采样
Convolutional Neural Network(CNN)
RCF network
building edge detection
stepped upsampling
