The Cyborg Astrobiologist: Matching of Prior Textures by Image Compression for Geological Mapping and Novelty Detection
This work addresses geological exploration for astrobiological research, providing incremental improvements in autonomy for robotic rovers and assistance for human astronauts.
The paper tackles the problem of geological mapping and novelty detection in astrobiology by using an image-compression technique for texture comparison, achieving 91% accuracy in similarity detection and 64% accuracy in novelty detection in field tests with heterogeneous rock types.
(abridged) We describe an image-comparison technique of Heidemann and Ritter that uses image compression, and is capable of: (i) detecting novel textures in a series of images, as well as of: (ii) alerting the user to the similarity of a new image to a previously-observed texture. This image-comparison technique has been implemented and tested using our Astrobiology Phone-cam system, which employs Bluetooth communication to send images to a local laptop server in the field for the image-compression analysis. We tested the system in a field site displaying a heterogeneous suite of sandstones, limestones, mudstones and coalbeds. Some of the rocks are partly covered with lichen. The image-matching procedure of this system performed very well with data obtained through our field test, grouping all images of yellow lichens together and grouping all images of a coal bed together, and giving a 91% accuracy for similarity detection. Such similarity detection could be employed to make maps of different geological units. The novelty-detection performance of our system was also rather good (a 64% accuracy). Such novelty detection may become valuable in searching for new geological units, which could be of astrobiological interest. The image-comparison technique is an unsupervised technique that is not capable of directly classifying an image as containing a particular geological feature; labeling of such geological features is done post facto by human geologists associated with this study, for the purpose of analyzing the system's performance. By providing more advanced capabilities for similarity detection and novelty detection, this image-compression technique could be useful in giving more scientific autonomy to robotic planetary rovers, and in assisting human astronauts in their geological exploration and assessment.