Automatic detection of impact craters on Mars using a modified adaboosting method

The accurate recognition of impact craters is important to analyze and understand the relative dating of Martian surface. Since manually identifying small craters in a deluge of high-resolution Martian images is a tremendous task, a robust automatic detection algorithm of the crater is needed, but subject to lots of uncertainties and low successful detection rates. In this paper, a modified adaboosting approach is developed to detect small size craters on Mars. First, we construct a dual-threshold weak classifier based on the characteristics of the feature value distribution instead of the single threshold classifier. Second, we adjust the criterion of updating weights in the process of training. The small craters on Mars are autamatically detected based on the modified algorithm using the images from the High Resolution Stereo Camera (HRSC) onboard Mars Express with a resolution of 12.5 m/pixel. A high threshold with 0.85 is determined, and the true detection rate of small size craters on Mars is improved by almost 10% when compared to the original method. The true detection rate can be obtained as high as 85% with only 10% false detection rate. Therefore, the modified adaboosting method has greatly improved the detecting performance of the crater and reduced the detection time.