Abstract:
This paper makes the first attempt to utilize long short-term memory (LSTM) for classification of solar radio spectrum. A solar radio spectrum is a gray-scale image repre...Show MoreMetadata
Abstract:
This paper makes the first attempt to utilize long short-term memory (LSTM) for classification of solar radio spectrum. A solar radio spectrum is a gray-scale image representing solar radio radiation over multiple frequency channels and in a short time period. The vertical and horizontal dimensions of a spectrum correspond to frequency channel and time, respectively. Intrinsically, time dependence exists between columns of a spectrum, which indicates the slowly varying process of solar radio radiation. Thus, a spectrum can be treated as a time sequence instead of a general image losing its time information. Inspired by the big success of LSTM for time sequence processing, e.g., natural language recognition, it is reasonably believed that treating a spectrum as a time sequence would benefit its classification via LSTM. Thus, LSTM is employed to explore the sequential relations and interactions within a spectrum being treated as a time sequence. As such, LSTM learns the sequential properties and generates the representation of solar radio spectrum for classification. The experimental results demonstrate that LSTM can well capture the characteristics of solar radio spectrum, and thus achieves better classification accuracy.
Date of Conference: 10-14 July 2017
Date Added to IEEE Xplore: 07 September 2017
ISBN Information: