结合空间像素模板和多类AdaBoost的高分影像分类

Classification of High Resolution Remote Sensing Image by Combining Spatially Correlated Pixels Template and Multi-class AdaBoost

高分辨率遥感影像分类是遥感图像理解的基本问题之一,也是许多其他遥感应用的前提。为解决目前基于像素分类方法空间关系描述不足的问题,该文利用空间像素模板构建像素间的空间关系,并结合多类AdaBoost算法实现高分辨率遥感影像的分类。首先利用过滤式特征选择方法自动生成空间像素模板,进而构建考虑空间关系的多维特征向量,最后利用基于指数损失函数的多类AdaBoost方法对多维特征进行分类。对不同场景影像开展实验,结果表明,该文方法利用空间像素模板引入空间信息,可有效实现高分辨率遥感影像分类。与其他方法相比,分类精度显著提高(约20%),能够更好地区分光谱相似地物,同时分类结果"椒盐效应"大大降低,具有良好的空间一致性。

The classification of high resolution remote sensing image is one of the basic problems in remote sensing image understanding.In order to solve the problem of the classification of remote sensing image based on pixel,a method is proposed in this paper by combining the spatially correlated pixels template with multi-class AdaBoost to obtain the classification of high resolution remote sensing image.Firstly,a specific form of spatially correlated pixels is generated by using the feature selection based on filter.Then,the feature vectors are formed using the spatially correlated pixels template,which contain the spatial information.Finally,a multi-class AdaBoost algorithm based on the exponential loss function is used to classify these feature vectors.Experimental results show that the proposed method is used to classify the high resolution remote sensing images effectively,which builds the spatial information with the spatial pixel template.Compared with other methods,the accuracy of classification results of the proposed method is higher (about 20%).Meanwhile,the result of proposed method has better spatial consistency with lower effect of pepper and salt.The confusing geo-objects which have similar spectral characteristics can be distinguished well by the proposed method.