论文部分内容阅读
为更有效地实现复杂场景中的多类目标同时检测,本文提出了一种基于多核学习算法进行目标检测的框架。该方法由特征提取和模型训练2个阶段组成。特征提取阶段,引入了多尺度下的点特征、表观特征同时对多类目标进行综合描述;模型训练阶段,分别采用加权相加和相乘2种方法将提取的各个基础特征组合起来,在支持向量机的框架下对各特征所代表的基础核权重进行学习。将训练所得的分类器结合滑动窗搜索技术对遥感图像进行目标检测实验,结果表明,与传统单核支持向量机相比,准确率更高。
In order to achieve simultaneous detection of multiple targets in complex scenes more effectively, this paper proposes a framework for target detection based on multi-core learning algorithm. The method consists of two stages: feature extraction and model training. At the stage of feature extraction, point features and appearance features at multi-scale are introduced. At the same time, multi-class objects are described synthetically. During the model training stage, the basic features extracted are combined by weighted summation and multiplication. Under the framework of support vector machine, the basic nuclear weights represented by each feature are studied. The training classifier is combined with the sliding window search technique to test the remote sensing images. The results show that the accuracy of the proposed method is higher than that of traditional single-kernel SVM.