基于最优特征选择与支持向量机的钱塘江涌潮检测算法  

Optimal feature selection and support vector machines-based algorithm for detection of tidal bore in Qiantangjing River

作　　者：高鹏[1] 王瑞荣[1] 王培力[1] 

机构地区：[1]杭州电子科技大学生命信息与仪器工程学院,浙江杭州310018

出　　处：《水利水电技术》2017年第1期40-45,共6页Water Resources and Hydropower Engineering

基　　金：国家自然科学基金项目(61374005);浙江省自然科学基金项目(LY14F030022)

摘　　要：以设计一种全新的背景模型算法为目标,依据图像像素特征之间的差异,使用高斯核函数获取像素特征的概率密度,统计不同区间的密度估计值,从特征池中选择适合的像素特征组成特征模板图,以输入视频流中对应位置的像素特征值作为输入量,使用支持向量机训练固定数目的像素特征值,对比分离出前景和背景。该方法应用于钱塘江涌潮检测的结果表明,F-measure值均在65%以上,鲁棒性较强;支持向量机方法选择径向核函数的识别率超过90%,运算速度较高。该方法能减少水面波动的干扰,具有较高的精度,可为河流动力特征描述提供重要工具。By taking the design of a fully new background model algorithm as the objective,a feature template is composed by the suitable pixel features selected from the relevant feature pool with the probability densities obtained by Gaussian kernel function through counting the density estimated values among various intervals,and then both the foreground and the background can be separated through the comparison with a fixed number of the pixel eigenvalues trained by the support vector machines by taking the pixel eigenvalues at the corresponding positions in the input video-streams as the input. The result from the application of this method to the detection of the tidal bore in Qiangtangjiang River shows that all the F-measure values are over 65% with stronger robustness and the recognition rate for selecting radial basis function with support vector machine is over 90% with higher computing speed,and then it can reduce the disturbance from the fluctuation of water surface with a high accuracy,thus can provide an important tool for description of the dynamic characteristics of the river.

关 键 词：最优特征选择 支持向量机 背景建模 运动目标检测 涌潮检测 钱塘江 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]