层叠式SOM神经网络星图识别算法及FPGA验证

Star Pattern Recognition Algorithm and FPGA Verification of Stacked SOM Neural Network

为了对空间天文望远镜精细导星仪获得的星图完成识别,提出一种层叠式自组织映射(SOM)神经网络算法模型,将该模型在硬件中实现星特征矢量匹配算法。首先,针对精细导星仪的特点详细介绍了导航星库的建立、星特征矢量的构建和筛选方法;其次,建立层叠式SOM神经网络模型,对其权值进行在线训练;最后,设计算法离线运行硬件电路并将其在FPGA中实现。仿真与测试结果表明,基于层叠式自组织神经网络的星图识别算法识别率高、抗噪声能力强、识别速度快。星点位置噪声为0.648?,星等噪声为0.18视星等条件下星图识别成功率在80%以上,新算法在FPGA中运行速度是PC机上传统三角形法的100倍。对精细导星仪星图识别算法的优化设计提供了合理可行的参考依据。

Astacked SOM neural network algorithm model is proposed for star identification,which is needed by fine guidance sensors,and the model is implemented in hardware to implement the star feature vector matching algorithm.Firstly,the method of establishment of the navigation star library and the construction of the star feature vector are introduced in detail.Then,the stacked SOM neural network model is put forward and the weights are trained online.Finally,the hardware circuits of the new algorithm are designed and implemented in the FPGA.The simulation and test results show that the star pattern recognition algorithm based on the stacked SOM neural network has a high recognition rate,strong anti-noise capacity,and fast recognition speed.The star map recognition success rate under harsh conditions is more than 80%,and the recognition speed is 100 times more than that of traditional methods that run in PCs.This research provides a reference for the design of fine guidance sensor star image recognition algorithms.