一种基于Hough变换和神经网络的分层类星体识别方法

A STRATIFIED APPROACH FOR QUASAR RECOGNITION BASED ON HOUGH TRANSFORM AND NEURAL NETWORK

类星体是宇宙中最明亮、密集的天体。它产生于宇宙诞生早期,具有重要的研究价值。观测的类星体光谱由于红移现象,光谱向长波方向偏移,因此识别类星体观测光谱中的发射线和确定类星体的红移是类星体识别的主要目标。类星体光谱固有的高噪声和观测光谱特性,给类星体识别带来很大困难。一般来说基于规则的直接匹配方法在类星体识别中效果不佳。本文介绍一种神经网络和Hough变换(HT)结合的类星体自动识别方法。该方法具有简单、快速、高效、鲁棒性强和通用性强等特点。

Quasar Objects (QSOs) are detectable at very large distance, with broad, red-shifted emission lines, strong ultraviolet and strong time variability of the optical light. QSOs play an important role in the research of the universe. The main purposes of quasar recognition are to identify the emission peaks in an observable quasar spectrum and to determine the observable quasar's redshift value. Due to the inherent extremely noisy characteristics of quasar spectrums and the limitation of observable conditions, automatic quasar recognition is a hard problem to tackle, and the commonly used direct matching approaches based on rules are ineffective. This paper introduces a stratified approach based on Hough transform and neural network which is shown to be simple, efficient, robust and easy to generalize.