论文部分内容阅读
根据经典径向基函数(RBF)神经网络的优势,结合星图模式样本集的特点,设计了一种适合星图模式样本的网络训练算法。从提取星图模式入手,引入三角剖分理论,将可能出现在同一视场内的恒星以三角形的形式连接起来,提取连接的角距作为星图模式,建立了具有完备性、平移旋转不变性的星图模式样本集。然后,利用RBF神经网络做星图识别,研究顺序训练方法和批量训练方法,总结多种经典算法的优缺点,并设计了一种训练方法。通过实验证明了该种方法较其他经典算法更为适合学习星图模式样本。最后,给出RBF神经网络相关的训练数据,并通过模拟星图软件获得若干模拟星图作为观测样本,利用已经训练好的神经网络进行识别。试验结果表明,测试网络能够正确识别这些星图。
According to the advantages of classical radial basis function (RBF) neural network and combining with the characteristics of the star pattern pattern set, a network training algorithm suitable for the star pattern pattern is designed. Starting with the extraction of the star pattern, the triangulation theory is introduced to connect the stars that may appear in the same field of view in the form of triangles, and to extract the connection angle as the star pattern, and to establish a complete, translatory rotation invariant Star Pattern Pattern Sample Set. Then, using RBF neural network to do star image recognition, the sequential training method and batch training method are studied, the advantages and disadvantages of many classical algorithms are summarized, and a training method is designed. Experiments show that this method is more suitable for studying star pattern than other classical algorithms. Finally, RBF neural network related training data are given, and several simulated star charts are obtained as observation samples through the simulated star chart software, and the trained neural network is used for identification. The test results show that the test network can correctly identify these stars.