Detecting patterns of a technological intelligence in remotely sensed imagery

A statistical classification approach for detecting artificial patterns in satellite imagery such as those produced by a technological intelligence, and its application to the search for non-natural features of possible extraterrestrial origin on planetary surfaces is presented. Statistics of natural terrestrial backgrounds (fractal textures, drainage patterns, tectonic features, etc.) and artificial features (e.g., roads, cities, vehicles, archaeological ruins) are computed over a set of terrestrial training images. Images are represented by measurements of their fractal dimension, fractal model fit, anisotropy, and rectilinearity. The likelihood ratio (conditional probability of a measurement given it is artificial divided by the conditional probability of a measurement given it is natural) is used as an index for assessing the artificiality of an unknown image relative to the training set. A classification accuracy of 85% is achieved over a training set of terrestrial images. The statistics of the training set are then extended to assess a number of enigmatic lunar and planetary features. Results suggest that certain areas on our moon and on Mars appear to be artificial by comparison with terrestrial features.