基于并联卷积神经网络的无人机遥感影像建筑区域测量

Building Area Measurement of Uav Remote Sensing Image Based on Parallel Convolution Neural Network

摘要:

无人机遥感影像覆盖范围广，难以区分建筑区域与背景区域，所以研究基于并联卷积神经网络的无人机遥感影像建筑区域测量方法。获取无人机遥感影像，通过静态输出、图像融合、去雾等环节完成遥感影像预处理。构建并联卷积神经网络，通过网络训练传播提取无人机遥感影像建筑区域边缘特征，经过特征匹配实现无人机遥感影像中建筑区域识别，结合面积计算结果获取建筑区域的测量结果。经过精度性能测试实验得出结论，在有雾和无雾环境下所提方法与传统区域测量方法相比的建筑区域测量误差分别降低了0.505km2和0.305km2，说明该方法的测量结果可靠性更高。

Abstract:

The coverage of drone remote sensing images is wide, making it difficult to distinguish between building areas and background areas. Therefore, a measurement method for building areas using drone remote sensing images based on parallel convolutional neural networks is studied. Obtain drone remote sensing images and complete remote sensing image preprocessing through static output, image fusion, defogging, and other processes. Construct a parallel convolutional neural network, extract the edge features of the building area in drone remote sensing images through network training propagation, and recognize the building area in drone remote sensing images through feature matching. Combined with the area calculation results, obtain the measurement results of the building area. After precision performance testing experiments, it was concluded that the proposed method reduced the measurement errors of building areas by 0.505km2 and 0.305km2 respectively compared to traditional area measurement methods in foggy and non foggy environments, indicating that the measurement results of this method are more reliable.