基于BP神经网络的船舶号灯识别模型与仿真

Simulation of Ship Lights Recognition Model Based on Back Propagation Neural Network

建立基于误差反传神经网络的船舶号灯智能识别模型,在众多的号灯识别参数中进行优化分析,确定了能见度、号灯亮度、背景亮度和眩光4个重要输入参数;利用这4个参数,基于误差反传神经网络对船舶号灯的可识别性进行建模和仿真,比较利用Levenberg-Marquart(L-M)、动量梯度下降、变学习率动量梯度下降和弹性反向传播等学习算法建立的误差反传神经网络模型,并确定L-M算法具有最优结果.通过号灯识别的仿真结果表明,识别结果与航海专家评估的结果一致.本模型实现了复杂光环境下船舶号灯可识别性的预报和影响因素分析,对保障船舶的夜航安全有着重要意义.

An intelligent identification model for ship lights was established on the basis of back propagation（BP） neutral network. Parameters of visibility, average luminance of ship lights, average background luminance and glare were selected among various influence parameters, and were set as inputs of the ship lights identification model. The BP neural network model were established by means of Levenberg-Marquart algorithm, adaptive learning rate momentum （ L-M ） algorith.m, momentum gradient descent gradient descent algorithm, and elastic back propagation algorithm respectively, and the L-M algorithm was selected as the optimal method and utilized for establishing the identification model. Results of ship lights identification simulation showed that the simulated results of the identification model were in accordance with the assessment results of experts in navigational field. The model realizes the prediction and the influence analysis of the identifiability of ship lights under complicated navigational luminous environments, and is of great importance for navigation safety at night.