A computer program undergoing development detects patterns that may differ in size but are otherwise similar to a specified pattern. Con- ceived to enable the automated recognition of features in images of planets and asteroids acquired by exploratory spacecraft, the program can also be used for scale-invariant recog- nition of patterns in other applications. The program requires no advance knowledge or mathematical modeling of a pattern to be recognized; instead, the program trains itself on one or more examples of a pattern provided by the user. The program synthesizes virtual examples by resampling the user-provided example(s) at different pixel spacings. The result of the resampling is a set of continuously scalable detectors, which can be regarded as implementing an extension of matched filtering (also known as template matching in the computervision and pattern-recognition literature), which was developed in the early 1940’s for radar and communication applications. The program has shown promise in tests on images of terrain of several astronomical bodies. For example, in the case of images of bowl-floored Lunar craters wider than 4 pixels, the program ex- hibited an 80-percent probability of detection and a 12-percent false-alarm rate.
This program was written by Michael C. Burl and Timothy Stough of Caltech for NASA’s Jet Propulsion Laboratory.
This software is available for commercial licensing. Please contact Don Hart of the California Institute of Technology at (818) 393- 3425. Refer to NPO-30269.
This Brief includes a Technical Support Package (TSP).
Software Recognizes Similar Patterns of Different Sizes
(reference NPO-30269) is currently available for download from the TSP library.
Don't have an account?
Overview
The document presents a technical support package from NASA's Jet Propulsion Laboratory (JPL) detailing a novel computer program developed by Michael C. Burl and Timothy M. Stough. This program is designed to automatically detect, size, and classify geological features in planetary and asteroidal data sets, addressing the challenges posed by the increasing volume of data returned by NASA spacecraft. The primary focus of the program is the detection of craters, which are significant for understanding the relative ages of planetary surfaces.
The program employs machine learning and computer vision techniques, allowing it to train itself on one or more user-provided examples of a specific pattern, such as craters. Unlike traditional methods that rely on domain-specific knowledge or mathematical modeling, this algorithm synthesizes virtual examples by resampling the provided examples at different pixel spacings. This results in continuously scalable detectors that extend the concept of matched filtering, a technique originally developed for radar and communications.
The effectiveness of the program has been tested on various celestial bodies, including the Moon, Europa, Callisto, and asteroid 433 Eros. Notably, it achieved an 80% probability of detecting craters larger than four pixels in diameter on the Lunar Maria, with a false alarm rate of 12%. However, the algorithm faced challenges when applied to more complex surfaces, such as those on Europa and Callisto, where the variability in crater appearance complicated detection.
In addition to crater detection, the program has shown potential for identifying other geological features, such as blocks and boulders on asteroids and lava cones on Earth, by simply adapting the training examples used. This adaptability highlights the program's versatility and its potential applications beyond planetary science.
The document also mentions that the software is available for commercial licensing, indicating its potential for broader use in various fields requiring pattern recognition capabilities. Overall, this innovative program represents a significant advancement in automated image analysis, promising to enhance scientific understanding of planetary geology and facilitate new discoveries in space exploration.
