Recognition of volcanoes on Venus using correlation methods

摘要

Radar images of 95% of the surface of Venus have been obtained by the Magellan spacecraft at resolutions of 100–300 m. The surface area covered is three times the total land-mass area of the Earth; this corresponds to a data volume of about 1011 bytes. A large population of volcanoes has been observed in this data set. Measurements of these features are essential for a full understanding of Venusian geology. The scale of the task, however, precludes the use of manual methods to make these measurements. An algorithm for the automated location and counting of these volcanoes is therefore being developed. The noisy nature of the data makes it appropriate to use correlation-based techniques to recognize the features. A least-squares-error template matching algorithm has been implemented, which includes local DC removal and contrast normalization. Preliminary experimental results from running the algorithm on Magellan data are presented, along with the corresponding measurements of expert human observers. Because there is no ground truth information for Venus, it has also been necessary to undertake a control experiment, using simulated radar images of artificial terrain. The results of this experiment are also included, and their implications for the calibration of both human and automated measurements are discussed.