基于小波变换与卷积神经网络的自动搜索X射线天文暴事件的方法 postprint

Abstract: We present an automatic peak-finding method based on the wavelet transform and convolutional neural network to identify the peaks of light curves, and it helps to rapidly discover potential transient sources from massive data. Here we use the real observation data of the HXMT as our experiment data. The experiments make comparisons with the wavelet transform peak-finding method, which is commonly used for peak-finding problems of spectral or chemical spectral. The results show that the method improves the peak-finding precision rate and the F1 value by 42% and 41.3%, and reduces the false alarm rate by 42.1%, while the peak-finding time only increases by 0.18 s. Therefore it can solve the problem of light curve peak-finding by combining the wavelet transform and convolutional neural network.