基于BP神经网络的高分辨率遥感影像分类

Classification of High-resolution Image Based on Spectral and Textural Features by Back-propagation Neural Network

为了能有效地从高分辨率遥感影像中提取地物信息,本文通过影像的光谱和纹理特征,利用BP神经网络算法进行影像分类研究。首先提取分类所需的光谱和纹理特征源,然后根据影像和地物特征,建立BP神经网络,用于样本训练和分类处理,实现地物分类。为验证该方法的可靠性,以2006年11月获取的成都平原某区域的Quickbird影像为实验数据,进行高分辨率遥感影像的地物分类实验。实验结果表明,结合影像光谱和纹理特征的BP神经网络分类算法,不仅可以有效保证BP神经网络分类训练的稳定性和收敛速度,还能达到较高的分类精度。

To extract the object-level information from the high-resolution satellite imagery,this paper presents a non-parametric classification approach termed back-propagation（BP） neural network.Both spectral and textural features extracted from high-resolution images are jointly used in the BP solution for the purpose of implementing classification.The experiments are performed using the multi-spectral（resolution 2.44 m） and panchromatic（resolution 0.61 m） images acquired by the satellite Quickbird in November of 2006 around the certain plain area of Chengdu,Sichuan.The quality of BP classification is assessed by checking two indicators,i.e.,overall accuracy and Kappa coefficient.The testing results show that the joint use of spectral and textural features can take high classification accuracy,as well as ensuring the training stability and convergence rate.