10:45 AM - 12:15 PM
[PCG18-P12] Study on the Automatic Event Classification System using an Machine Learning Inference on FPGA
Keywords:FPGA, Machine Learning, Plasma Wave
Automatic event classification using a machine learning model is an effective approach for improving the intelligent processing on spacecraft. In general, GPUs (Graphical Processing Units) are used for machine learning and inference. However, when we implement these functions in a spacecraft, we need to use alternative devices due to power consumption and thermal design constraints. In this study, we evaluate the performance of machine learning inference functions implemented in the TUL PYNQ-Z2 board (based on Xilinx Zynq-7000) to aiming for automatic event classification observed in space.
Binarized Neural Network (BNN) is a neural network in which some (or all) of trainable parameter are represented by binary values. The BNN-PYNQ package, which helps implementation of BNN on PYNQ-Z2, provides a simple LFC (fully-connected network) model and CNV (convolutional network) model. We found that the inference time using the programable logic (FPGA) on the PYNQ-Z2 was approximately 2070 times faster than that using the processing system (CPU) on the PYNQ-Z2, and approximately 1.6 times faster than that using an NVIDIA Tesla V100 (GPU) device.
We proposed a machine learning approach for classifying plasma wave spectrogram observed by the PWE aboard Arase into three classes (broadband electric noise, chorus, and hiss). We trained a fully-connected machine leaning model using binarized images (50 x 50 pixels) of the observed electric field spectrograms, and achieved an inference accuracy over 96%.
Binarized Neural Network (BNN) is a neural network in which some (or all) of trainable parameter are represented by binary values. The BNN-PYNQ package, which helps implementation of BNN on PYNQ-Z2, provides a simple LFC (fully-connected network) model and CNV (convolutional network) model. We found that the inference time using the programable logic (FPGA) on the PYNQ-Z2 was approximately 2070 times faster than that using the processing system (CPU) on the PYNQ-Z2, and approximately 1.6 times faster than that using an NVIDIA Tesla V100 (GPU) device.
We proposed a machine learning approach for classifying plasma wave spectrogram observed by the PWE aboard Arase into three classes (broadband electric noise, chorus, and hiss). We trained a fully-connected machine leaning model using binarized images (50 x 50 pixels) of the observed electric field spectrograms, and achieved an inference accuracy over 96%.