利用BP神经网络提取TM影像水体

常用的水体提取方法需要设定阈值,而阈值的选择通常需要反复实验才能确定。为克服这一不足,本文提出了一种基于BP神经网络和光谱特征的水体自动提取方法。首先在TM影像中提取水体样本光谱信息、谱间关系、归一化差异水体指数以及缨帽变换(K-T变换)的第三分量(TC3)特征,然后将这4个特征作为BP神经网络的训练输入参数,最后利用训练好的网络对水体进行提取。对长沙市区TM影像进行水体提取,发现该方法组合了其他方法的优点,在不设置阈值的情况下,得到了更好的水体提取效果。

Water body extracting from TM image based on BPNN

Traditional method of water body extraction needs to set the right thresholds,and threshold selection usually requires repeated experiments to determine.In order to overcome the shortage,a novel method to extract water body was proposed in this paper base on spectral characteristics and BP neural network.First,the spectral information,the Spectrum-photometric,the normalized difference water index,the third component(TC3)of Tasseled Cap(K-T transform)characteristics from the water body of TM image was extracted,and then the four characteristics was taken as trained parameters of the BPNN,finally the trained network was used to extract water.Through the experiments of water body extraction form TM images of Changsha,it was found that this method could combine the advantages of other methods to get a better water extraction effect without setting threshold value.