论文部分内容阅读
在遥感影像飞机目标检测中,由于目标训练样本的局限性及遥感影像尺寸较大带来的复杂性,直接使用级联式Adaboost算法会产生较多的虚警,因此需要采取一定手段去除虚假目标。近来,Hough森林算法因其简单而有效的表现在目标检测中有较多的应用。然而,直接使用Hough投票机制应用于遥感影像全图,时间耗费比较大。因此,本文将两种算法结合起来,首先由级联Adaboost算法检测出候选目标区域,然后通过改进的Hough森林算法对这些候选目标区进行二次筛选。在二次筛选中,由于初检已经确定目标的可能位置,因此不需要对全图的图像块进行位置投票,只需用对该区域作目标存在可能性评价,降低了时间消耗。试验表明,本文方法不仅能很好地去除虚假目标,同时也保证了检测时间的有效性。
In the remote sensing image plane target detection, because of the limitation of the target training sample and the complexity caused by the large size of the remote sensing image, the direct use of the cascade Adaboost algorithm will generate more false alarms, therefore some measures need to be taken to remove the false target . Recently, Hough forest algorithm has more applications in target detection because of its simple and effective performance. However, using the Hough voting mechanism directly for the entire image of the remote sensing image can take a long time. Therefore, this paper combines two algorithms. Firstly, the candidate target regions are detected by the cascade Adaboost algorithm, and then the candidate target regions are screened by the improved Hough forest algorithm. In the secondary screening, since the preliminary examination has determined the possible location of the target, there is no need to vote the position of the image block of the whole image, and only the possibility of using the target area for the region is evaluated, reducing the time consumption. Experiments show that this method can not only remove the false target well, but also ensure the validity of the detection time.