Morphometry, Votes-Analysis and Calibration Improvements of Crater Detection Algorithms Based on Edge Detectors and Radon/Hough Transform
暂无分享,去创建一个
Six previously implemented Crater Detection Algorithms (CDAs) were improved using morphometry measurements (some new and some improved), votes-analysis and calibration. The result s were analyzed using the Framework for Evaluation of CDAs (FECDA). Introduction: CDAs’ applications range from dating planetary surfaces [1] to advanced statistical an lysis [2]. CDAs are an important subject of recent sc ientific research [3-10]. Additional overview of CDArelated literature is given in [11]. In our previou s work on CDAs, we proposed: (1) implementation based on standard gradient edge detectors and Radon/Hough transform [12-13]; and (2) several CDAs’ specific i mprovements of the initial implementation [14-15]. Methods: The new methods and the improvements of the methods from previous work [15] are as follo ws. New morphometry measurements. Automated measurements of crater rim and central peak are per formed using 2-D crater profile. Associate volumes (profile surfaces) are compared to the volume of a crater itself. The circular consistency [14] of these features is additionally measured. The experiments sho w that the higher volumes and circularity mean the hi gher probability that detected feature is a crater. Improved morphometry measurements. An automated detection of radial range where a crater is p reserved is performed. The purpose is to check the ci rcular consistency [14] only on the preserved part of a crater. In combination with the appropriate weight factor which depends on radial range, this increases o v rall performance. In order to improve tuning [14], a dditionally is performed: (1) an automated detection o f smaller circular features inside a larger one; (2) an automated depth/diameter measurement; (3) usage of more combinations and previously computed values. Circularity analysis of parameter space. This new method evaluates parameter space in the same way as the circularity consistency evaluates 3-D crater sh ape [14]. The experiments show that the circularity of v tes in parameter space is higher at craters’ centers th n in the centers of false detections. Therefore, this ne w measurement is used to improve overall performance. Calibration of resulting catalogue. This new method multiplies probability that the detected fea ture is a crater with calibration factor. This factor de pends only on a detected radius and increases with the in crease of a radius. This partially compensates diff erences in morphology between small and large craters . Other changes. The radius range is increased from 5~10 to 5~28 pixels while the optimal gradient is t he same as in the previous work [15]. Numerous other parameters are also optimized. Results: The obtained results are shown in Table 1. The analysis using F-ROC and detected edges are shown in Fig. 1. For evaluation of the results, fro m FECDA [11] the following were used: (1) 1/64° MOLA data; (2) the older GT-17582 catalogue [11] (and not the newer GT-57633 catalogue [16]) so that the results can be compared with the previous work [15]; and (3) Topolyzer application. Conclusion: As the results show, CDAs were significantly improved using morphometry measurements, votes-analysis and calibration. Only ~3.24% of the craters from the GT-17582 catalogue are still undetected, mostly highly eroded (ghost) craters. Table 1: Used gradient edge detectors and obtained results before and after modifications described in this pa per.