摘要
研究对航空高光谱赤潮图像进行快速、准确地识别问题。传统单一BP网络在解决航空遥感高光谱图像大数据量问题时网络结构复杂、训练时间长、识别速度低。针对大数据量快速分析的需求,提出了一种集成结构的遗传模块化神经网络的航空高光谱赤潮识别方法。首先对学习任务进行划分,通过i/j分类方法组建训练样本集。然后,每个子神经网络模块仅仅针对特定小区域进行训练,为了避免传统BP网络在选择网络结构及参数时仅凭经验或反复大量实验的缺点,采用自适应遗传算法对网络结构及参数进行了优化。最后,通过模糊隶属度将分块学习的模块化神经网络进行集成。实验证明这种可以快速有效地对赤潮进行监测,为设计提供依据。
Study the quick and effective monitoring of the airborne hyper-spectral red tide.In order to resolve the problem of complicated structure and long training time when traditional BP network deals with the massive airborne hyper-spectral remote data,this paper presented a genetic modular neural network to monitor the red tide.This method divided the massive data learning assignment into small learning assignments.Firstly,it divided the learning assignment and formed the learning samples by i/j.And then,every sub-neural-network was trained with the corresponding samples,whose structure and training parameters were determined by an Adaptive Genetic Algorithm（AGA）.Lastly,the every sub-neural-network was combined into a neural network by the combination module.The results of the experiments show that this method can monitor red tide rapidly and effectively.
出处
《计算机仿真》
CSCD
北大核心
2012年第5期258-261,共4页
Computer Simulation
基金
山东省高等学校科技计划项目(J09LG03)
