摘要
针对无人机倾斜测量的建筑特征提取与分类精度低的问题,提出一种结合SSD(Single Shot MultiBox Detector)算法和最小显著性差异法(Least Significant Difference,LSD)算法的建筑特征提取与分类方法。首先将SSD算法模型原始卷积层建立为目标特征层,并采用空洞卷积底部采样方式结合深层特征与浅层特征,实现SSD算法改进;然后采用改进SSD算法对含目标建筑物的图像进行筛选;最后利用LSD算法对筛选出的图像进行分类识别,以此实现建筑特征提取与分类。仿真结果表明,所提方法可有效提高无人机倾斜测量的建筑特征提取与分类识别精度,相较于Radon和PPHT算法模型,所提方法对建筑物中心识别准确率和组识别准确率最高,分别达到94.75%和83.49%,可实现1:500的航空摄影测量大比例尺地形图绘制。
Aiming at the problem of building feature extraction and low classification accuracy of UAV tilt measurement,a building feature extraction and classification method combining SSD(Single Shot MultiBox Detector)algorithm and minimum significant difference method(Least Significant Difference,LSD)algorithm is proposed.Firstly,the original convolution layer of SSD algorithm model is established as the target feature layer,and then the improvement of deep features and shallow features adopts SSD algorithm;finally,the LSD algorithm is used to realize building feature extraction and classification.Simulation results show that the proposed method can effectively improve the drone tilt measurement building characteristic extraction and classification identification accuracy,compared with Radon and PPHT algorithm model,the proposed method of building center recognition accuracy and group recognition accuracy,reached 94.75%and 83.49%respectively,can achieve 1:500 aerial photography large scale topographic map.
作者
周波
王法景
唐桂彬
ZHOU Bo;WANG Fajing;TANG Guibin(Yangling Vocational Technical College,Xianyang Shanxi 712100,China)
出处
《自动化与仪器仪表》
2023年第7期38-41,46,共5页
Automation & Instrumentation
基金
2021年度陕西高等职业教育教学改革研究项目(21GY006)
陕西省教育厅2022年度一般专项科研计划项目(22JK0626)。