基于SVM的决策融合鱼类识别方法  

Decision fusion fish identification using SVM and its experimental study

作　　者：杜伟东[1,2] 李海森[1,2] 魏玉阔[1,2] 徐超[1,2] 

机构地区：[1]哈尔滨工程大学水声技术重点实验室,黑龙江哈尔滨150001 [2]哈尔滨工程大学水声工程学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2015年第5期623-627,共5页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目(41306038);水声技术重点实验室基金资助项目(9140C200105120C2001)

摘　　要：为解决基于声学散射数据的高精度鱼类识别问题,提出一种基于SVM的多方位声散射数据决策层融合的鱼类识别方法。利用小波包变换(WPT)和离散余弦变换(DCT)方法对多方位声散射数据进行特征提取,并进行特征降维处理。然后采用SVM分类器对每个方位提取的特征做出多次决策,并输出最终识别结果。采用3种不同鱼类作为研究对象,设计了可靠的获取多方位声散射数据的实验方案,给出不同方位数量条件下,基于WPT和DCT特征量的识别率。理论分析及实验数据处理结果表明,随着方位数量的增加,总体识别率呈升高的趋势,基于SVM的多方位声散射数据决策层融合方法可以有效提高识别率至90%以上。In order to meet the requirement of high precision for fish identification based on acoustic scattering data, an SVM-based multi-azimuth decision fusion method for acoustic scattering data is proposed in this study. Firstly, the wavelet packet transform （ WPT） and discrete cosine transform （ DCT） methods are used to extract features from multi-azimuth acoustic scattering data and the extracted features are processed for a dimension-reduction. Sec-ondly, SVM classifiers are employed to make decisions for multiple times based on the features of each azimuth. Fi-nally, the identification result is figured out as the ultimate output. In the experiment, three fishes are selected and the reliable scheme is designed to obtain multi-azimuth acoustic scattering data. The identification rates are demon-strated for each case of different azimuth numbers using the WPT and DCT methods. The theoretical analysis and processing results indicated that the overall identification rates show a rising trend as the azimuth number increases. The SVM-based multi-azimuth decision fusion method for acoustic scattering data can increase the identification rate to 90%.

关 键 词：鱼类识别 多方位 决策融合 支持向量机 小波包变换 离散余弦变换 

分 类 号：S932.4[农业科学—渔业资源] P745[农业科学—水产科学]