Automated Development of Feature Extraction Tools for Planetary Science Image Datasets

Introduction: There is more image data available to planetary scientists now than at any time in history. The problem now is how to best make use of it. It is impractical to analyze such large datasets manually, while development of handwritten feature extraction tools is expensive and laborious. This project explores the use of machine learning techniques to develop feature extraction algorithms for the Mars Orbiter Camera (MOC) narrow angle dataset using the Los Alamos GENIE machine learning software. GENIE, the GENetic Image Exploitation package, uses a genetic algorithm to assemble feature extraction algorithms from low-level image operators. Each algorithm is evaluated against user-provided training data, and the most accurate ones are allowed to "reproduce" to build the next generation. The algorithm population evolves until it converges to a solution or specified level of accuracy. Mars Global Surveyor (MGS) [1] has been on orbit around Mars since 1997. It carries many scientific instruments including the Mars Orbiter Camera (MOC) [2]. The narrow angle dataset, the focus of this study, provides imagery with a spatial resolution of order 3 meters/pixel and is sensitive to light in a broad visi-ble/near-infrared spectral range (0.50mm – 0.90mm). Since arriving at Mars, MOC has taken over 112,000 images, which have been used to study various planetary processes. Craters were selected as the feature of interest for this study because they are a discrete, easily recognizable feature that can be used to derive much information about a surface [3-4]. GENIE: Los Alamos National Laboratory's GENIE software [5] uses techniques from genetic algorithms (GA) [6-8] and genetic programming (GP) [9] to construct spatio-spectral feature extraction algorithms for multi-spectral remotely sensed imagery. Both the algorithm structure and the parameters of the individual image processing steps are learned by the system. GENIE has been described at length elsewhere [10-12], so we will only present a brief description here. In particular, the present work explores using GENIE on panchromatic imagery [13-14]. GENIE follows the paradigm of genetic programming: a population of candidate image-processing algorithms is randomly generated from a collection of low-level image processing operators. The fitness of each individual is assessed from its performance on training data provided by the human user via a graphi