摘要
为了提高违章建筑识别精度,提出一种基于U-net网络的无人机影像违章建筑自动识别方法。利用无人机获取建筑影像数据,通过裁剪、几何校正、重叠区域校正完成无人机影像的预处理。在U-net网络的下采样部分,加载金字塔池化单元,并通过深度可分离卷积改进U-net网络。通过学习训练样本完成改进U-net网络的训练,将测试样本输入该网络。将网络输出结果与历史地形图对比,实现违章建筑的自动识别。实验结果表明,该方法具有更高的识别精度,可实现无人机影像的几何校正,拟合误差最小,校正效果最优,违章建筑识别准确率高。
To improve the identification accuracy of illegal buildings,an automatic identification method based on U-net network is proposed.UAV is used to obtain building image data,and the preprocessing of UAV image is completed through cropping,geometric correction,and overlapping area correction.Load the pyramid pooling module in the downsampling part of the U-net network and improve the U-net network by depthwise separable convolution.The training of the improved U-net network is completed by learning the training samples,and the test samples are input into the network.Compare the network output results with historical topographic maps to realize automatic identification of illegal buildings.The results show that this method has higher recognition accuracy,and can realize geometric correction of UAV images,with the smallest fitting error,the best correction effect,and high recognition accuracy of illegal buildings.
作者
马佳斌
周振兴
MA Jiabin;ZHOU Zhenxing(Yangtze River Delta(Jiaxing)Urban-Rural Construction and Design Group Co.,Ltd.,Jiaxing 314050,China)
出处
《江西测绘》
2022年第3期13-16,共4页
JIANGXI CEHUI
