摘要
船舶夜间航行需要通过船舶号灯识别其型号、编号及航道等信息。为了提高船舶识别的准确率,本文建立了基于BP神经网络的船舶号灯识别模型。通过将信号灯亮度、投射距离、背景亮度等作为输入参数,并将已有数据作为输入,训练BP神经网络,修正网络内部神经节点个数及反馈函数。分别使用已有的L-M函数、动量梯度下降、无监督学习等反馈函数训练和验证,L-M反馈函数可得到精度最高的神经网络模型。通过使用BP神经网络模型对船舶号灯进行识别,提高了船舶夜间航行的安全性。
The ships have to be identified with their types, serial number and channel information through navigation light in night travelling. In order to raise the identification accuracy, the BP neural network identification on navigation light model is built. Through setting the light brightness, project distance, background brightness as the input parameters, we train the neural network with the correct identification data, and modify the node amounts and feedback function. Based on experiments on L-M function, momentum gradient descent function and unsupervised learning, we come to conclusion that the LM feedback function can provide the best accuracy for the network model. The security of night travel for shipments are highly raised through identifying the navigation light through BP neural network.
出处
《舰船科学技术》
北大核心
2017年第2X期175-177,共3页
Ship Science and Technology
基金
江西省普通本科高校中青年教师发展计划访问学者专项资金资助项目
校级教改课题资助项目(JY1503)
校级团队建设资助项目(JXTD1404)