为了对高维非线性的高光谱影像进行降维及信息提取,提出了高光谱影像核最小噪声分离变换(kernel minimumnoise fraction,KMNF)特征提取后利用BP神经网络分类的方法。以高光谱影像KMNF特征提取后的前几个特征分量作为BP神经网络的输入,进行BP神经网络分类,并与单独的高光谱影像BP神经网络分类进行比较。美国内华达州CU-PRITE矿区AVIRIS数据的实验结果表明,基于KMNF和BP神经网络的高光谱影像分类较单独BP神经网络分类总体精度及时间性能均得到提高。
To reduce dimension and extract information of hyperspectral images,a hyperspectral imaging BP neural network classification method after kernel minimum noise fraction(KMNF) transformation is presented.First few feature vectors of KMNF is regarded as BP neural network inputs and then hyperspectral images are classified by BP,and it is compared with single BP neural network classification.CUPRITE Nevada USA AVIRIS data experimental results show that hyperspectral image KMNF and BP neural network classification overall accuracy increases and performs quicker compared with single BP neural network classification.