Conveners
Deep Learning in On-Board Systems
- David Steenari (ESA)
The possibilities to observe and interact with any given spacecraft are naturally limited compared to ground-based systems due to a number of factors. These include but are not limited to the availability and bandwidth of their connection to ground, the availability of staff, communication latencies and power budgets.
While a minimum level of autonomy is required for every spacecraft, past...
Specific technological innovations are required to accomplish more ambitious commercial and scientific goals for space missions. One of the key areas of potential innovation is mission autonomy: an increased degree of on-board autonomy helps in implementing more effective mission operations. In particular, functionalities like event detection, autonomous planning and goal management, if...
Modern imaging sensors create frames of several megapixels at a high frame acquisition rate. These imaging sensors are placed onboard spaceborne platforms and can greatly enhance their capabilities. However, an industry trend towards smaller satellites - with smaller antennas, less power and worse pointing accuracy- leads also to an expectation that the downlink capability will remain well...
Satellite based spectroscopic observations produce massive volumes of data at very high velocities. Analysis of such observation via deep learning algorithms, e.g. for the precise estimation of the redshift associated with individual galaxies, requires massive floating point operations, generally performed on Graphics Processor Units (GPU). This form of processing is fast but requires...
Deep learning is enabling technology for many application like image processing, pattern recognition, objects classification and even autonomous space craft operations. But, there is a price to be paid, these methods are computationally intensive and require supercomputing resources - that can be challenging, especially on-board of a space craft. However, FPGA-accelerated computing is becoming...
Nanosatellites typically operate on a basis of scheduled, routine procedures, defined by users on the ground and dictated via pass uplinks. The development of machine learning algorithms, combined with constant advancement in the efficiency and capabilities of nanosatellite systems, has led to the point where artificial intelligence may be deployed on small satellites via low-power components...
We present the deep learning platform (N2D2) and the neural network hardware IPs (PNeuro and DNeuro) developed at CEA, which are specifically tailored to design and integrate deep networks in highly constrained embedded systems (using low power GPU, FPGA or ASIC). The software platform integrate database construction, data pre-processing, network building, benchmarking and optimized code...
We report on the ongoing hardware and software developments to implement cloud screening in-orbit and in real-time. We leverage a Vision Processing Unit to accelerate state-of-art artificial Intelligence algorithms applied on hyperspectral and thermal imaging data. The instrument on which the developments are implemented is HyperScout-2, second generation of a very compact hyperspectral system...
Recently, there is an increasing interest in the use of Artificial Neural Networks (ANN) in space, given the huge success of such algorithms in terrestrial applications. Industrial applications such as industrial robotics, security, unmanned autonomous vehicles and driverless cars have been driving the acceleration of such algorithms in embedded systems. In addition, even in data centres...
As we begin to acquire high-resolution video from space, the potential for new and disruptive satellite-based technologies grows. Real-time airport and road traffic monitoring, increased maritime domain awareness (MDA) through vessel detection and tracking, and timely warning systems for natural disasters are some of the possible avenues being explored. Even in the absence of object motion in...
