基于目标特征模型配置的面向对象检测方法研究

The Research of Object-Oriented Detection Method Based on Technique of Configurable Target Feature Model

针对目前遥感图像目标检测过程中过度提取特征容易导致样本特征维数过高,造成维数灾难的问题,提出一种基于目标特征模型配置的面向对象检测方法,该方法以特征的可分性为原则判断特征的真实有效性,针对不同的侦查目标,将真正有用的特征配置于该待检测的目标,达到快速、准确自动检测的目的。试验结果表明,该方法能在降低算法复杂度的同时提高目标检测的正确率,能节省样本训练所需的时间。

In the processing procedure of remote sensing image, the over extraction of target feature tends to increase the dimension of target feature, consequently leads to the problem of dimension disaster. Therefore, based on the separability of characteristic, the object can be automatically, accurately and rapidly identified by using the true and effective characteristic for detecting different objects. The experimental results show that the method proposed can reduce the algorithm complexity, promote the ratio of detection validity simultane- ously and shorten the sample training period as well.