一种基于自动特征学习的陨石坑区域检测算法

基于陨石坑的视觉导航技术成为一种新颖的高精度空间探测自主导航方式，如何从导航图像中精确地提取陨石坑区域是实现基于陨石坑视觉导航的首要条件。针对这一问题，根据陨石坑导航图像特点，提出了一种基于自动特征学习的陨石坑区域检测算法。首先，基于最大稳定极值区域检测算法提取陨石坑候选区域；其次，利用卷积神经网络（CNN）自动学习提取候选区域的特征；最后，通过支持向量机（SVM）实现候选区域的精确分类，得到真实的陨石坑区域。大量的仿真实验表明:与传统的基于人工特征的陨石坑区域检测算法相比，提出的基于自动特征学习的陨石坑区域检测算法具有更高的检测精度和更好的鲁棒性，在通用火星表面陨石坑数据集上，所提算法的F₁度量指标较于传统算法高出8%，可以广泛地应用于基于陨石坑的视觉导航算法中的陨石坑区域提取，为基于陨石坑视觉导航算法提供精确的导航路标输入。 

A crater region detection algorithm based on automatic feature learning

The crater-based navigation technology has been become a novel and precise autonomous navigation method in space exploration, and how to extract the crater regions from the crater navigation image is the essential condition of the crater-based navigation method. Accordingly, in this paper, we propose an algorithm for extracting crater regions via automatic feature learning. First, the candidate crater regions were obtained by the maximal stable external region method. Then, the features of these regions were automatically extracted by Convolutional Neural Network (CNN). Finally, the true crater regions were identified from all the candidate regions through Support Vector Machine (SVM) classifier. The experimental results demonstrate that the proposed algorithm can extract crater regions from the navigation image with higher accuracy and robustness than the traditional crater region detection algorithms based on the handcrafted features. The proposed algorithm obtains an F₁ score which is 8% higher than that of the traditional method on the standard Mars surface crater database, and can be applied in the crater detection of the crater-based visual navigation method to provide the precise navigation landmarks.