论文部分内容阅读
针对交通事件数据样本少,检测效率低的问题,将加权支持向量机引入到交通事件检测中,采用样本重要度加权法提高算法的检测率,根据识别误差确定样本重要度权值,建立了交通事件检测的样本重要度加权法支持向量机算法,最后应用实测数据对标准支持向量机算法、样本重要度加权法、样本数目加权法3种算法的检测效果进行测试。研究结果表明:样本数目加权法算法能够根据样本的好坏自适应确定样本重要度权值,提高了算法的鲁棒性;当负正样本比率减少时,3种算法的检测效果均变差,而对于同样的样本,标准支持向量机的检测率最低,样本重要度加权法的效果最好,加权算法的选择要依据样本的数量、分布不平衡以及识别目标而定;在交通事件检测中,为了提高检测率,选择样本重要度加权效果最好,在不同的样本不平衡率下,检测效果是不同的,不平衡率越严重,检测效果越差。
In order to reduce the number of traffic events and the low detection efficiency, weighted support vector machines are introduced into traffic incident detection. The sample importance weighted method is used to improve the detection rate of the algorithm. The importance of the samples is determined according to the identification error. Traffic is established Event-based sample importance weighting method support vector machine (SVM) algorithm is proposed. Finally, the measured data is used to test the detection results of the standard support vector machine algorithm, sample importance weighted method and sample number weighted method. The results show that the weighted number of samples method can adaptively determine the weight of sample importance according to the quality of the samples and improve the robustness of the algorithm. When the ratio of negative samples decreases, the detection results of the three algorithms become worse, For the same sample, the standard SVM has the lowest detection rate, the sample importance weighted method has the best effect, and the selection of weighted algorithm depends on the number of samples, the unbalanced distribution and the recognition target. In traffic incident detection, In order to improve the detection rate, the weighted value of sample importance is the best. Under different sample unbalance rates, the detection results are different. The more the unbalance rate is, the worse the detection effect is.