基于克隆优化的船舶号灯神经网络识别模型  

Neural network recognition model of ship lights based on clonal optimization

作　　者：朱金善[1] 孙立成[2] 胡江强[1] 何庆华[1] 

机构地区：[1]大连海事大学航海学院,辽宁大连116026 [2]中国船级社,北京100007

出　　处：《大连海事大学学报》2015年第2期41-45,共5页Journal of Dalian Maritime University

基　　金：辽宁省自然科学基金资助项目(2014025008);中央高校基本科研业务费专项资金资助项目(3132014028)

摘　　要：针对复杂光环境下船舶号灯识别模型的高维、强非线性及影响因素复杂等特性,提出一种基于克隆选择优化算法的BP神经网络识别模型.通过对影响因素的筛选确定BP神经网络的输入,将号灯识别码作为网络的输出确定BP神经网络模型.采用免疫克隆选择优化算法,确定网络层数和各层节点数目,结合灵敏度分析法选择非线性寻优的方向和尺度,以减少BP神经网络的迭代次数,提高搜索效率.通过对海上夜航时拍摄的一些实景照片进行学习和识别的仿真,验证了所建立的船舶号灯识别模型的有效性.Aiming at the features of ship lights recognition model under complex light environment such as strong nonlin- earity, high dimension and complex environmental disturb- ances, a ship lights recognition model based on BP neural network was proposed. The relevant factors were selected as network inputs, and the identification code was set as network output to make up BP neural network model. Immune clonal selection optimization algorithm was adopted to decide the net- work layers and the number of units in each layer, by combi- ning with sensitivity analysis method to search the direction and scale of nonlinear optimization to reduce the number of iterations the BP neural network and improve the search effi- ciency. Based on the spot photoes of ship lights, simulations of ship lights recognition were conducted by using the improved neural recognition model. Simulation results demonstrate the efficiency of the proposed ship lights recognition model.

关 键 词：船舶号灯 识别模型 克隆优化 神经网络 

分 类 号：U665.16[交通运输工程—船舶及航道工程]