Fast Onboard Texture Analysis for Planetary Exploration

Texture analysis holds significant promise for improving the science return of remote planetary exploration. Communications with exploration spacecraft suffer significant latency and bandwidth constraints; onboard image understanding can summarize large datasets and select representative images for transmission. We present image texture descriptors that satisfy the strict computational requirements of flight processors. We use the integral image transform to produce pixel-level texture features. A decision tree using integer arithmetic computes pixel classifications far faster than texture analysis techniques requiring convolutions of steerable filter banks. A case study characterizes sedimentology in Mars Exploration Rover microscopic images. Here, the decision trees produce image descriptors that correlate strongly with geologists’ classifications of sediment types in the Gusev crater region. This in turn permits a kernelized regression model that predicts soil compositions using image texture without explicitly identifying individual particles.