Sound Spectrum Detection Using Deep Learning

Abstract

Spectrum detection which is called spectrum sensing, aims to periodically monitor a specific frequency band and determine the presence or absence of the signals. There are several techniques which are used to spectrum detection such as energy detection and matched filter detection [10]. Most of the study on spectrum detection is used these techniques. But there is no work on spectrum detecting that uses either deep learning or machine learning methods. So for this experiment, Convolutional Neural Network (CNN) which is one of the most common deep neural network architectures is used. The main aim of this study is to binary classification of sound, that is, determine the presence or absence of the signals by CNN. In the literature, generally deep neural networks are used for sound detection. One of the reasons we choose this method is that CNN is convenient to use for sound data. Our architecture has five convolutional layers and two fully connected layers. For this task we used 20,000 data and 16,000 of them were used for training, 2,000 of them were used validation and the rest of them were used for test.