Research of Contour Detection for Disk Resolved Objects Based on ResNet and Window-Contour Detection Method
-
摘要: 在空间图像的面元天体进行测量的过程中, 边缘提取是一个重要的部分. 使用传统算子如Roberts算子等进行检测时, 容易受到噪声和天体表面细节的影响. 提出一种基于残差神经网络(ResNet)的边缘提取方法, 可基于足够的数据集提取更精确的边缘. 同时, 也提出了一种具有鲁棒性的, 不依赖于数据集与计算资源的方法, 命名为窗口-边缘检测法. 对New Horizons LORRI图像的实验结果表明, ResNet 能够检测到更多的边缘像素点, 而窗口-边缘检测法则能够去除更多的非边缘像素点.
-
关键词:
- 边缘检测 /
- 残差神经网络 /
- New Horizons LORRI图像 /
- 面元天体
Abstract: During the process of astrometry of the disk resolved objects, contour detection acts as an important part. When using the traditional operators such as Roberts operator, the result is easily affected by the noise and interior details. A contour detection method based on ResNet is proposed, which can extract more accurate contours based on sufficient dataset. Meanwhile, a robust method named window-contour detection is proposed, which is not dependent on the dataset and computing resource. The experimental results on the New Horizons LORRI frames show that ResNet can detect more contour pixels, while window-contour detection method can exclude more non-contour pixels. -
-
计量
- 文章访问数: 165
- HTML全文浏览量: 23
- PDF下载量: 14
-
被引次数:
0(来源:Crossref)
0(来源:其他)
下载:
