Design of Convolutional Extreme Learning Machines for Vision-Based Navigation Around Small Bodies
This work addresses the need for faster and more efficient neural network training in space imagery applications, particularly for navigation around small bodies, though it is incremental as it builds on existing CELM methods.
The authors tackled the problem of computationally expensive training in deep learning for vision-based navigation by designing convolutional extreme learning machines (CELMs), which achieved comparable performance with dramatically reduced training time, enabling efficient architecture exploration.
Deep learning architectures such as convolutional neural networks are the standard in computer vision for image processing tasks. Their accuracy however often comes at the cost of long and computationally expensive training, the need for large annotated datasets, and extensive hyper-parameter searches. On the other hand, a different method known as convolutional extreme learning machine has shown the potential to perform equally with a dramatic decrease in training time. Space imagery, especially about small bodies, could be well suited for this method. In this work, convolutional extreme learning machine architectures are designed and tested against their deep-learning counterparts. Because of the relatively fast training time of the former, convolutional extreme learning machine architectures enable efficient exploration of the architecture design space, which would have been impractical with the latter, introducing a methodology for an efficient design of a neural network architecture for computer vision tasks. Also, the coupling between the image processing method and labeling strategy is investigated and demonstrated to play a major role when considering vision-based navigation around small bodies.