154.com皇冠
“科研成果快报”
1. 标题
· 一种基于邻近效应的暗目标提取方法—以裂缝提取为例
· A Dark Target Detection Method Based on the Adjacency Effect: A Case Study on Crack Detection
2. 成果信息
· Li Yu, Yugang Tian*, Wei Wu. (2019). A Dark Target Detection Method Based on the Adjacency Effect: A Case Study on Crack Detection. Sensors. 19. https://doi.org/10.3390/s19122829
· 资助项目:National Key Research and Development Program (Grant No.2018YFB1004600).
3. 成果团队成员
于丽,第一作者,硕士生,154.com皇冠,研究方向:图像处理。
田玉刚,通讯作者,副教授,154.com皇冠,研究方向:城市与环境遥感,图像处理与模式识别。
吴蔚,硕士,中国能源建设集团广东省电力设计研究院有限公司。
4. 成果介绍
1. 研究意义
随着遥感影像分辨率的不断增加和应用定量化要求。邻近效应成为了一个不可忽视的问题。当前关于邻近效应的研究大都是关于如何消除其影响,关于邻近效应应用的文章寥寥无几。本研究关注于邻近效应带来的暗目标位置和灰度对应的特征,并尝试使用这些特征辅助暗目标检测,设计了一种高低阈值检测策略和一种自适应阈值选取方法,通过与canny-morphology、SWT算法的精度和算法复杂度对比,表明,利用邻近效应进行暗目标检测是可行的,且本文算法复杂度较低。
2. 方法设计
1) 一种高低阈值暗目标检测策略
利用邻近效应带来的暗目标灰度与暗目标位置对应的特征,提出一种结合高低阈值提取结果的暗目标检测策略。其示意图如下:
Figure 1. Concept map of Low-High threshold detection strategy. And (a) is the conceptual map consists of noise, dark target and bright background. (b) is the conceptual map of detected result using a low threshold. (c) is the conceptual map of detected result using a high threshold. (d) is the conceptual map of rough detection result using the low-high threshold detection strategy.
2)一种自适应高低阈值获取方法
主要利用特征如下:
3 结果展示
以混凝土护坡的伸缩缝检测为例,canny-morphology 和SWT 算法作对比,结果对比图如下:
Figure 2. Detection accuracy for expansion joints of all images. And (a) is the Precision curves using the proposed method, the canny_morphology method, and the SWT Algorithm. (b) is the Recall curves using the proposed method, the canny_morphology method, and the SWT Algorithm. (c) is the F-measure curves of using the proposed method, the canny_morphology method, and the SWT Algorithm.
Figure 3. Four typical partial examples of expansion joint detection. Columns from left to right are the original images, manually drawn sketches, rough detection results using the proposed method, accurate detection results using the proposed method, detection results using the canny-morphology method, rough detection results using the SWT algorithm and accurate detection results using the SWT algorithm.
4 创新点
1) 首次将邻近效应应用于暗目标检测。
2) 提出了一种高低阈值检测策略和自适应阈值选取方法。
3) 将文本检测的SWT算法引入到伸缩缝检测,并与本文提出的方法进行对比。