Neural Networks Train Lasers to Spot Space Junk

A set of algorithms for laser ranging telescopes, developed by scientists from the Chinese Academy of Surveying and Mapping and the Liaoning Technical University, is improving the success rate of space debris detection in Earth’s orbit by significantly improving the pointing accuracy of the telescope.

After more than half a century of space activity, collisions between jettisoned engines and disintegrated satellites have created a celestial scrapheap that can be difficult for spacecraft to evade. Scientists have developed space junk identification systems, but it has proven tricky to pinpoint the swift, small specks of space litter.

Beijing Fangshan Satellite Laser Observatory. Courtesy of Beijing Fangshan Satellite Laser Observatory.

Laser ranging technology uses the reflection from objects to measure their distance. Space debris, because of its small size, poor prediction accuracy, and lack of surface reflection, falls into the category of a “noncooperative target,” causing the pointing position of the telescope to deviate from the actual position of the debris, and thus impeding the laser ranging system from accurately detecting the debris. Previous methods improved the ability of laser ranging telescopes to pinpoint debris, but only to a 1-km level.

Researcher Tianming Ma and his colleagues trained a back propagation neural network to recognize space debris. The genetic algorithm and Levenberg-Marquardt algorithm optimized the neural network’s thresholds for recognition of space debris, while ensuring that the network wasn’t too sensitive and could be trained on localized areas of space. The team demonstrated the improved accuracy of its approach by testing it against three traditional methods — the mount model, the spherical harmonic function model, and the basic parameter model — at the Beijing Fangshen laser range telescope station.

The observation data of 95 stars was used to solve the algorithm coefficients from each of the models. Then, the pointing accuracy of the telescope was verified by the detection results of 22 stars. The new pointing correction algorithms proved to be the most accurate method, as well as being easy to operate and being able to perform well in real time.

“After improving the pointing accuracy of the telescope through a neural network, space debris with a cross-sectional area of 1 meter squared and a distance of 1500 kilometers can be detected,” Ma said.

The back propagation neural network model optimized by the genetic algorithm and Levenberg–Marquardt algorithm demonstrated a pointing accuracy of 3.42 in. in the azimuth and 2.44 in. in the pitch. When the telescope corrected by this model was used to detect space debris, the results showed that the pointing accuracy of the telescope probably increased to nine times in the azimuth and three times in the pitch, compared to its performance before correction.

The results of the new study show that the back propagation neural network model optimized by the algorithms greatly increased the pointing accuracy of the telescope, improving the success rate of space debris detection. Ma plans to refine the method further. “Obtaining the precise orbit of space debris can provide effective help for the safe operation of spacecraft in orbit,” he said.

The research was published in Journal of Laser Applications (www.doi.org/10.2351/1.5110748).