摘要
无人机拍摄的低空遥感图像比普通遥感图像有更高的分辨率和更多细节,但它们也容易受到各种降质因素,尤其是光照的影响,给提取目标轮廓的任务带来了巨大的挑战。较强的降质因子可导致低空遥感图像部分轮廓信息的丢失,导致传统轮廓检测方法无法检测到完整的轮廓。为解决这一问题,以光伏板边缘检测为例,提出了一种基于先验模型优化的轮廓目标提取算法。该算法根据目标轮廓的几何形状先验生成轮廓模板,然后匹配初步检测到的不完整轮廓的和轮廓模板的关键点,最后根据相应的关键点的坐标对轮廓模板进行优化,得到完整的轮廓。实验结果表明,本算法相较于Canny边缘检测算法和基于深度学习的HED算法能更好地克服降质因素的影响,检测到完整的轮廓。
Low-altitude remote sensing images taken by drones have higher resolution and more details than ordinary remote sensing images,but they are also susceptible to various degradation factors,especially illuminations,which bring a great challenge to the task of extracting contour targets.Stronger degradation factors can result in the loss of the contour information in some parts of the sensing image,making it impossible for the traditional contour detection methods to detect complete contours.To deal with this problem,this paper uses edge detection of photovoltaic panels as an example and proposes a contour target extraction algorithm based on geometric contour detection.The algorithm generates a contour template according to the geometry of the target contour,then matches the key points of the detected incomplete contour and of the contour template,and finally optimize the contour template according to the corresponding key points to obtain a complete contour.We use the Canny edge detection algorithm,the deep-learning based HED algorithm and the proposed method to extract the contours of the low-altitude remote sensing images affected by illumination and evaluate the performance of each method.The experimental results verify the effectiveness of the proposed method.
作者
兰传琳
方佩章
何楚
LAN Chuanlin;FANG Peizhang;HE Chu(College of Electronic Information, Wuhan University, Wuhan 430072, China)
出处
《电视技术》
2019年第1期5-10,65,共7页
Video Engineering
关键词
轮廓检测
低空遥感图像
先验模型优化
降质因素
关键点匹配
contour detection
low-altitude remote sensing image
prior model optimization
degrading factor
key points matching
