| Summary: | 碩士 === 逢甲大學 === 通訊工程學系 === 107 === Convolutional neural networks are important network architectures for deep learning, and they have been widely used for pattern recognition with high accuracy. However, most of the data in reality are mostly Non-Euclidean Data. The general convolutional neural network cannot handle such data. In recent years, Graph Convolutional Neural Networks [10] [11] have been proposed for non-Euclidean data, as shown in the figure. The biggest advantage of the graph convolutional neural network is that it can analyze the data type of the graph form, allowing the input to be of any size, reducing the difficulty of long-term signal analysis, and not being inferior to the signal structured data dimension. This study proposes a new deep learning network based on graph convolutional neural network for continuous wavelet transform [12][13] and discrete square wave short-time Fourier transform [14] [15] The converted image data to prove the effectiveness of the graph convolutional neural network for the signal spectrogram to apply more different fields.
|
|---|
