Small Lunar Crater Detection From LROC NAC Using Statistically Constrained Path Morphologies and Unsupervised Discriminate Correlation Filters
Yaqiong Wang,Huan Xie,Qian Huang,Xiongfeng Yan,Shijie Liu,Zhen Ye,Chao Wang,Xiong Xu,Sicong Liu,Yanmin Jin,Xiaohua Tong
DOI: https://doi.org/10.1109/tgrs.2024.3401874
IF: 8.2
2024-06-04
IEEE Transactions on Geoscience and Remote Sensing
Abstract:The investigation of small lunar craters holds scientific and engineering significance. This article presents a novel method for detecting small lunar craters. It consists of three stages: seed detection, candidate crater detection, and crater evaluation. First, crater seeds are identified through morphological operations as pixels with the highest local gradient and specific gradient direction. Second, the optimal scale for each seed is estimated based on the maximum response of the established phase congruency maximum moment (PCMM) scale space. For detecting very small craters, a crater detector called statistical morphological constraint path-sets (SMPS), which leverages image spatial domain features, is proposed. It configures the image as a weighted directed graph, using path-sets centered on seeds to flexibly detect highlights and shadow regions of craters. For detecting craters with larger optimal scale, another crater detector named structural consistency constrained multipaths (SCMP) is proposed, utilizing the frequency phase features. The core idea of SCMP is to configure the phase feature type (PFT) map with the optimal scale as a directed graph. Centered on the seed, the multipath operator is designed to detect craters. Unsupervised discriminative correlation filters (UDCFs) are trained with histogram of oriented gradients (HOG) features from images or PFT maps to validate candidate craters. The results indicate that for images with a resolution of 0.5–2 m/pixel, the proposed method demonstrates good detection performance for small craters with diameters of less than 5, 5–10 m, and greater than 10 m, with an average detection rate (DR) of 0.89, 0.91, and 0.92, respectively.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics