一种基于BP网络的星图识别算法

A Star Pattern Identification Algorithm Based on BP Neural Network

摘要:

为了提高星图识别算法的抗噪性能,提出一种基于逆向传播(Back Propagation,BP)网络的识别算法。该算法通过将星图转换成“0”、“1”和“2”的网格矩阵,提取行列数值和星数形成匹配向量,利用多个BP识别子网进行训练完成匹配识别。通过仿真试验得出以下结论:对星等位置和星等添加噪声,当位置噪声标准偏差为2像素时,BP网络算法的识别率和识别时间相对传统栅格算法分别提高2%和60ms,对噪声有较强抗干扰能力,表明BP网络算法具有更快的识别速度。

Abstract:

In order to improve the anti-noise property of grid algorithm, an identification algorithm based on back propagation (BP) network was proposed. It extracts the number of rows and columns to form matching vector by transfroming the star pattern into a grid matrix of "0", "1" and "2". Then multiple BP recognition subnets were used for training to complete the matching recognition. Simulation results indicate that, when the standard deviation of position noise is 2 pixels, the recognition rate and the recognition time of BP network algorithm is improved by 2% and 60ms compared to grid algorithm after adding noise to both star locations and star magnitude. This algorithm based on BP shows strong anti-noise ability and faster identification speed.