基于决策树SVM分类器的感兴趣区域定位方法

感兴趣区域定位是提取目标特征,进行目标识别与跟踪等后续处理的重要基础.由于大尺寸遥感图像的光谱特性和目标形状均很复杂,通常采用的基于光谱特征的分割方法和基于边缘的区域生长技术不合适,从模式分类角度考虑遥感图像中感兴趣区域快速定位问题,提出一种基于决策二叉树支持向量机的纹理分类方法,将分类器分布在各个结点上,构成了多类支持向量机,减少了分类器数量和重复训练样本的数量.在SPOT图像上的实验结果表明,该方法实现感兴趣区域的快速定位有较高的分类正确率.

An Approach for ROI Fast Detection Based on Decision Tree SVM Classifier

Detecting regions of interest(ROIs) is the base step for object's feature extraction,identification and tracking.The segmentation algorithms based on optical characteristics and the region growing method based on edges are not suitable, because optical characteristics and object's shape are complex in remote sensing imagery with big size.So a texture classification method based on support vector machines(SVMs) decision binary tree is proposed.This method distributes classifier to each nodes which constitutes multi-class SVM.It can reduce the number of SVM classifier and duplicate training samples.Furthermore,the experiments are done on SPOT scenes.The results show that this method is adaptable,has high accuracy and speed.