论文部分内容阅读
为了从高光谱遥感影像中高精度提取各种线形道路,提出了基于支持向量机(SVM)的道路特征快速提取算法,首先利用PCA对高光谱影像进行合理压缩,由SVM模式识别理论推导出该算法具有快速精确提取道路网信息的能力,针对高光谱遥感影像高信息量和道路网复杂度高的特点,提出基于1Vm(一对多算法)的多种道路SVM一次性高精度提取的多分类策略,在提高精度的同时,兼顾了道路特征识别的效率。研究结果表明:SVM对线状道路模式判别能力比常规方法有更强的优势,对小样本的道路识别效果更加明显,从遥感影像中不仅能准确地辨别出道路的线形特征,还能识别出其材质和类型;该算法能同时识别出多种道路,执行效率更高。
In order to extract all kinds of linear roads with high accuracy from hyperspectral remote sensing images, a fast feature extraction method based on Support Vector Machine (SVM) is proposed. Firstly, hyperspectral images are reasonably compressed by PCA, and the algorithm is derived from SVM pattern recognition theory Aiming at the high information capacity of hyperspectral remote sensing images and the high complexity of road network, this paper proposes a multi-classification strategy based on 1Vm (one-to-many algorithm) , At the same time improve the accuracy, taking into account the efficiency of the road feature recognition. The results show that the SVM has more superiority than the conventional method in discriminating the linear road mode, and the effect of road recognition is more obvious for small samples. The remote sensing image can not only accurately identify the linear features of the road, but also identify Its material and type; the algorithm can identify a variety of roads at the same time, the execution efficiency is higher.