摘要
提出一种不需要设定阈值的BP-ANN的分布式水体识别方法。利用水体样本的基本光谱信息,结合NDVI(归一化植被指数)、MNDWI(归一化差异水体指数)等特征对反向传播(back propagation,BP)神经网络进行训练;设计基于GNDWI和谱间关系的样本自动选择算法,通过实验选取合适的特征组;构建基于MapReduce的分布式BP神经网络水体识别模型。实验结果表明,该模型具有稳定的可扩展性,在保证识别精度的同时,提高水体遥感提取的速度和自动化程度。
Without setting the threshold, a distributed method based on BP-ANN to identify waterbody was proposed. As the in- put of the BP neural network, the basic spectral information of water samples was used. Simultaneously, characteristics inclu- ding NDVI and MNDWI index were combined to train the BP neural network. Based on the GNDWI and relationship between spectrums, a sample auto-selection algorithm was designed to gain high quality training samples. Through the experiments, the optimal feature combination was selected. A distributed BP model to abstract waterbody was built based on the MapReduce pro- gramming model. Experimental results show that this model has stable extensibility. Meanwhile, it can enhance the speed of waterbody abstraction and improve the accuracy and automation.
出处
《计算机工程与设计》
北大核心
2015年第8期2229-2233,2244,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(41261090)
新疆研究生科研创新基金项目(XJGRI2014033)