基于目标约束与谱空迭代的高光谱图像分类方法

Hyperspectral Image Classification Method Based on Targets Constraint and Spectral-Spatial Iteration

针对复杂背景像元影响高光谱分类精度的问题,将目标检测方法引入地物分类研究,提出了一种基于谱空特征迭代的高光谱图像分类方法,该方法通过将约束能量最小化设计了一种多目标约束的类别分类器(MTCC)。该分类器利用检测原理提取多类目标地物,有效地降低了复杂背景数据对分类精度的影响;同时为了解决光谱特征带来的过分类问题,方法中利用反馈式谱空融合方式强化空间增强信息在分类中的作用,以逐步提高分类精度。利用Purdue、Salinas和Pavia数据集进行实验,结果表明,所提方法的平均分类精度分别为98.09%、97.33%和84.68%,精确率分别为96.84%、95.32%和79.13%,与其他方法相比所提方法具有更高的泛化能力,实用性更强。

Aiming to solve the problem that the complex background pixels affect the hyperspectral classification accuracy,the object detection theory is introduced into the hyperspectral image classification domain,and a hyperspectral image classification method based on spectral-spatial feature iteration is proposed.A multi-target constrained classifier（MTCC）is designed by constrained energy minimization method.Based on the detection theory,the MTCC can effectively decrease the influence of complex background data on the classification accuracy.At the same time,to eliminate the over-classification problem caused by the spectral features,the method uses the feedback fusion of spectral-spatial to strengthen the spatial enhancement information so as to improve the classification accuracy gradually.The results of the experiments on the data sets of Purdue,Salinas and Pavia show that the average accuracies of the proposed methods are 98.09%,97.33% and 84.68% respectively,and the precisions of the proposed method are 96.84%,95.32% and 79.13%respectively.Compared to other algorithms,the proposed method has higher generalization ability and practicability.